|
8489
|
162
|
63
|
2026-04-14T06:48:52.610946+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149332610_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
-206300887034301537
|
3979031157576197803
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny...
|
8488
|
|
8490
|
163
|
51
|
2026-04-14T06:48:52.610930+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149332610_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"bounds":{"left":0.09375,"top":0.045138888,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"bounds":{"left":0.09375,"top":0.046527777,"width":0.003515625,"height":0.18680556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"bounds":{"left":0.1,"top":0.05625,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"bounds":{"left":0.1171875,"top":0.05625,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"bounds":{"left":0.1328125,"top":0.05625,"width":0.022265624,"height":0.022222223},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"bounds":{"left":0.13515624,"top":0.061805554,"width":0.017578125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
-206300887034301537
|
3979031157576197803
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny...
|
NULL
|
|
8494
|
162
|
66
|
2026-04-14T06:49:00.678440+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149340678_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-c github.com/jiminny/prophet/pull/468/changes#diff-cac29662cfcc94ab0d41bf55628c52944063d9b3d1c18f441442c3c0da7f4fde...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
-6737127715865063816
|
3709976328637656239
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9...
|
NULL
|
|
8495
|
163
|
53
|
2026-04-14T06:49:00.611598+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149340611_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-c github.com/jiminny/prophet/pull/468/changes#diff-cac29662cfcc94ab0d41bf55628c52944063d9b3d1c18f441442c3c0da7f4fde...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
0
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
-E4rcp0Bd4RvtuaTN2RY.json
-E4rcp0Bd4RvtuaTN2RY.json
7067cZ0Bd4RvtuaT1GTC.json
7067cZ0Bd4RvtuaT1GTC.json
7k6scZ0Bd4RvtuaT9GQV.json
7k6scZ0Bd4RvtuaT9GQV.json
807ncZ0Bd4RvtuaTumTs.json
807ncZ0Bd4RvtuaTumTs.json
8E6-cZ0Bd4RvtuaThmSx.json...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"bounds":{"left":0.10625,"top":0.065972224,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"bounds":{"left":0.10625,"top":0.068055555,"width":0.03515625,"height":0.07083333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"bounds":{"left":0.10625,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"bounds":{"left":0.1328125,"top":0.060416665,"width":0.0140625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"bounds":{"left":0.1546875,"top":0.050694443,"width":0.1046875,"height":0.014583333},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"bounds":{"left":0.1546875,"top":0.052083332,"width":0.1046875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"bounds":{"left":0.2625,"top":0.052083332,"width":0.003515625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"bounds":{"left":0.26601562,"top":0.052083332,"width":0.01015625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"bounds":{"left":0.1515625,"top":0.0625,"width":0.03984375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"bounds":{"left":0.15507813,"top":0.06736111,"width":0.0265625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"bounds":{"left":0.19648437,"top":0.065972224,"width":0.037890624,"height":0.0125},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"bounds":{"left":0.19648437,"top":0.06736111,"width":0.037890624,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"bounds":{"left":0.2359375,"top":0.06736111,"width":0.068359375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"bounds":{"left":0.3058594,"top":0.06458333,"width":0.021484375,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"bounds":{"left":0.30820313,"top":0.06736111,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"bounds":{"left":0.32890624,"top":0.06736111,"width":0.01015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"bounds":{"left":0.340625,"top":0.06458333,"width":0.1203125,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"bounds":{"left":0.34296876,"top":0.06736111,"width":0.115625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"bounds":{"left":0.4625,"top":0.0625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"0","depth":15,"bounds":{"left":0.7796875,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"bounds":{"left":0.7828125,"top":0.06111111,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"bounds":{"left":0.7867187,"top":0.06111111,"width":0.008984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"bounds":{"left":0.796875,"top":0.06111111,"width":0.015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"bounds":{"left":0.8222656,"top":0.056944445,"width":0.0546875,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"bounds":{"left":0.8335937,"top":0.06111111,"width":0.03984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"bounds":{"left":0.88007814,"top":0.056944445,"width":0.0453125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"bounds":{"left":0.88359374,"top":0.06111111,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"bounds":{"left":0.9007813,"top":0.06111111,"width":0.01484375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"bounds":{"left":0.9285156,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"bounds":{"left":0.94921875,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"bounds":{"left":0.96171874,"top":0.056944445,"width":0.0203125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"bounds":{"left":0.97148436,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"bounds":{"left":0.9746094,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"bounds":{"left":0.9777344,"top":0.06111111,"width":0.001953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"bounds":{"left":0.11953125,"top":0.09861111,"width":0.080078125,"height":0.020833334},"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"bounds":{"left":0.203125,"top":0.09791667,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"bounds":{"left":0.10664062,"top":0.13125,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"bounds":{"left":0.10664062,"top":0.13333334,"width":0.016796876,"height":0.045833334},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"bounds":{"left":0.12539062,"top":0.0,"width":0.17070313,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"bounds":{"left":0.12851563,"top":0.0,"width":0.033203125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"bounds":{"left":0.12851563,"top":0.0,"width":0.012890625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"-E4rcp0Bd4RvtuaTN2RY.json","depth":23,"bounds":{"left":0.13164063,"top":0.0013888889,"width":0.07421875,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"-E4rcp0Bd4RvtuaTN2RY.json","depth":24,"bounds":{"left":0.13164063,"top":0.0013888889,"width":0.07421875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"7067cZ0Bd4RvtuaT1GTC.json","depth":23,"bounds":{"left":0.13164063,"top":0.023611112,"width":0.07578125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"7067cZ0Bd4RvtuaT1GTC.json","depth":24,"bounds":{"left":0.13164063,"top":0.023611112,"width":0.07578125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"7k6scZ0Bd4RvtuaT9GQV.json","depth":23,"bounds":{"left":0.13164063,"top":0.045833334,"width":0.076171875,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"7k6scZ0Bd4RvtuaT9GQV.json","depth":24,"bounds":{"left":0.13164063,"top":0.045833334,"width":0.076171875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"807ncZ0Bd4RvtuaTumTs.json","depth":23,"bounds":{"left":0.13164063,"top":0.068055555,"width":0.07539062,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"807ncZ0Bd4RvtuaTumTs.json","depth":24,"bounds":{"left":0.13164063,"top":0.068055555,"width":0.07539062,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"8E6-cZ0Bd4RvtuaThmSx.json","depth":23,"bounds":{"left":0.13164063,"top":0.090277776,"width":0.076171875,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
975097796230276763
|
1404131123152039147
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
0
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
-E4rcp0Bd4RvtuaTN2RY.json
-E4rcp0Bd4RvtuaTN2RY.json
7067cZ0Bd4RvtuaT1GTC.json
7067cZ0Bd4RvtuaT1GTC.json
7k6scZ0Bd4RvtuaT9GQV.json
7k6scZ0Bd4RvtuaT9GQV.json
807ncZ0Bd4RvtuaTumTs.json
807ncZ0Bd4RvtuaTumTs.json
8E6-cZ0Bd4RvtuaThmSx.json...
|
8492
|
|
8517
|
163
|
64
|
2026-04-14T06:49:31.748601+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149371748_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-f github.com/jiminny/prophet/pull/468/changes#diff-fc8a653dec2cd8e3d632fa5d32e0abd4b22176ce7619cc5752c4efe9e53f083b...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/endpoints.py
src/eval/endpoints.py
src/eval/endpoints.py
Copy file name to clipboard
Expand all lines: src/eval/endpoints.py
Lines changed: 5 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:
172
known_keys
=
{
172
known_keys
=
{
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
173
"content"
,
"summary"
,
"themes"...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"bounds":{"left":0.10625,"top":0.0,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"bounds":{"left":0.10625,"top":0.0,"width":0.03515625,"height":0.07083333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"bounds":{"left":0.10625,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"bounds":{"left":0.1328125,"top":0.0,"width":0.0140625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"bounds":{"left":0.1546875,"top":0.0,"width":0.1046875,"height":0.014583333},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"bounds":{"left":0.1546875,"top":0.0,"width":0.1046875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"bounds":{"left":0.2625,"top":0.0,"width":0.003515625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"bounds":{"left":0.26601562,"top":0.0,"width":0.01015625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"bounds":{"left":0.1515625,"top":0.0,"width":0.03984375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"bounds":{"left":0.15507813,"top":0.0,"width":0.0265625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"bounds":{"left":0.19648437,"top":0.0,"width":0.037890624,"height":0.0125},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"bounds":{"left":0.19648437,"top":0.0,"width":0.037890624,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"bounds":{"left":0.2359375,"top":0.0,"width":0.068359375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"bounds":{"left":0.3058594,"top":0.0,"width":0.021484375,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"bounds":{"left":0.30820313,"top":0.0,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"bounds":{"left":0.32890624,"top":0.0,"width":0.01015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"bounds":{"left":0.340625,"top":0.0,"width":0.1203125,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"bounds":{"left":0.34296876,"top":0.0,"width":0.115625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"bounds":{"left":0.4625,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"bounds":{"left":0.78046876,"top":0.0,"width":0.00234375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"bounds":{"left":0.7828125,"top":0.0,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"bounds":{"left":0.7867187,"top":0.0,"width":0.008984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"bounds":{"left":0.796875,"top":0.0,"width":0.015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"bounds":{"left":0.8222656,"top":0.0,"width":0.0546875,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"bounds":{"left":0.8335937,"top":0.0,"width":0.03984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"bounds":{"left":0.88007814,"top":0.0,"width":0.0453125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"bounds":{"left":0.88359374,"top":0.0,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"bounds":{"left":0.9007813,"top":0.0,"width":0.01484375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"bounds":{"left":0.9285156,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"bounds":{"left":0.94921875,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"bounds":{"left":0.96171874,"top":0.0,"width":0.0203125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"bounds":{"left":0.97148436,"top":0.0,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"bounds":{"left":0.9746094,"top":0.0,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"bounds":{"left":0.9777344,"top":0.0,"width":0.001953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"bounds":{"left":0.11953125,"top":0.0125,"width":0.080078125,"height":0.020833334},"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"bounds":{"left":0.203125,"top":0.011805556,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"bounds":{"left":0.10664062,"top":0.045138888,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"bounds":{"left":0.10664062,"top":0.047222223,"width":0.016796876,"height":0.045833334},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"bounds":{"left":0.12539062,"top":0.050694443,"width":0.17070313,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"bounds":{"left":0.12851563,"top":0.072916664,"width":0.033203125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"bounds":{"left":0.12851563,"top":0.09513889,"width":0.012890625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"bounds":{"left":0.12539062,"top":0.11736111,"width":0.01953125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"bounds":{"left":0.12851563,"top":0.13958333,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"bounds":{"left":0.12851563,"top":0.12638889,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"bounds":{"left":0.12851563,"top":0.14861111,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"bounds":{"left":0.12851563,"top":0.14861111,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"bounds":{"left":0.12539062,"top":0.17152777,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"bounds":{"left":0.12539062,"top":0.17152777,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"bounds":{"left":0.23203126,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"src/eval/endpoints.py","depth":15,"bounds":{"left":0.24453124,"top":0.0,"width":0.059375,"height":0.014583333},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXLink","text":"src/eval/endpoints.py","depth":16,"bounds":{"left":0.24453124,"top":0.0,"width":0.059375,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"src/eval/endpoints.py","depth":18,"bounds":{"left":0.24453124,"top":0.0,"width":0.059375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy file name to clipboard","depth":15,"bounds":{"left":0.30703124,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Expand all lines: src/eval/endpoints.py","depth":15,"bounds":{"left":0.31796876,"top":0.0,"width":0.0109375,"height":0.019444445},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 5 additions & 0 deletions","depth":15,"bounds":{"left":0.890625,"top":0.0020833334,"width":0.0234375,"height":0.099305555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXCheckBox","text":"Not Viewed","depth":14,"bounds":{"left":0.91914064,"top":0.0,"width":0.03125,"height":0.019444445},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Viewed","depth":16,"bounds":{"left":0.93046874,"top":0.0,"width":0.01640625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Comment on this file","depth":14,"bounds":{"left":0.95351565,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"More options","depth":14,"bounds":{"left":0.9675781,"top":0.0,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Original file line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Original file line","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line change","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"172","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"known_keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"=","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"{","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"172","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"known_keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"=","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"{","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"173","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"content\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"summary\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"themes\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"action_items\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"173","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"content\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"summary\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"themes\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
5656235918028717126
|
1685749034226822627
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/endpoints.py
src/eval/endpoints.py
src/eval/endpoints.py
Copy file name to clipboard
Expand all lines: src/eval/endpoints.py
Lines changed: 5 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:
172
known_keys
=
{
172
known_keys
=
{
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
173
"content"
,
"summary"
,
"themes"...
|
NULL
|
|
8519
|
162
|
78
|
2026-04-14T06:49:37.376827+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149377376_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-f github.com/jiminny/prophet/pull/468/changes#diff-fc8a653dec2cd8e3d632fa5d32e0abd4b22176ce7619cc5752c4efe9e53f083b...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/endpoints.py
src/eval/endpoints.py
src/eval/endpoints.py
Copy file name to clipboard
Expand all lines: src/eval/endpoints.py
Lines changed: 5 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:
172
known_keys
=
{
172
known_keys
=
{
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
174
"salesRep"
,
"customer"
,
174
"salesRep"
,
"customer"
,
175
+
"ai_activity_type"
,
"ai_activity_type_reasoning"
,
175
}
176
}
176
is_empty
=
known_keys
.
isdisjoint
(
new_data
.
keys
())...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"src/eval/endpoints.py","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXLink","text":"src/eval/endpoints.py","depth":16,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"src/eval/endpoints.py","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy file name to clipboard","depth":15,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Expand all lines: src/eval/endpoints.py","depth":15,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 5 additions & 0 deletions","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXCheckBox","text":"Not Viewed","depth":14,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Viewed","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Comment on this file","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"More options","depth":14,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Original file line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Original file line","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line change","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"172","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"known_keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"=","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"{","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"172","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"known_keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"=","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"{","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"173","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"content\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"summary\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"themes\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"action_items\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"173","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"content\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"summary\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"themes\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"action_items\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"174","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"salesRep\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"customer\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"174","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"salesRep\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"customer\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"175","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"ai_activity_type\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"\"ai_activity_type_reasoning\"","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"175","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"}","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"176","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"}","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"176","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"is_empty","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"=","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"known_keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"isdisjoint","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"new_data","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"keys","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"())","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
6204677564107301541
|
1686866138040643015
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/endpoints.py
src/eval/endpoints.py
src/eval/endpoints.py
Copy file name to clipboard
Expand all lines: src/eval/endpoints.py
Lines changed: 5 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -172,6 +172,7 @@ async def process_entry(idx: int, entry: Dict[str, Any]) -> None:
172
known_keys
=
{
172
known_keys
=
{
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
173
"content"
,
"summary"
,
"themes"
,
"action_items"
,
174
"salesRep"
,
"customer"
,
174
"salesRep"
,
"customer"
,
175
+
"ai_activity_type"
,
"ai_activity_type_reasoning"
,
175
}
176
}
176
is_empty
=
known_keys
.
isdisjoint
(
new_data
.
keys
())...
|
NULL
|
|
8521
|
163
|
66
|
2026-04-14T06:49:40.788505+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149380788_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-f github.com/jiminny/prophet/pull/468/changes#diff-fc8a653dec2cd8e3d632fa5d32e0abd4b22176ce7619cc5752c4efe9e53f083b...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import
AiActivityTypeResponse
12...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"bounds":{"left":0.10625,"top":0.06527778,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"bounds":{"left":0.10625,"top":0.06736111,"width":0.03515625,"height":0.07083333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"bounds":{"left":0.10625,"top":0.05625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"bounds":{"left":0.1328125,"top":0.059722222,"width":0.0140625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"bounds":{"left":0.1546875,"top":0.05,"width":0.1046875,"height":0.014583333},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"bounds":{"left":0.1546875,"top":0.05138889,"width":0.1046875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"bounds":{"left":0.2625,"top":0.05138889,"width":0.003515625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"bounds":{"left":0.26601562,"top":0.05138889,"width":0.01015625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"bounds":{"left":0.1515625,"top":0.061805554,"width":0.03984375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"bounds":{"left":0.15507813,"top":0.06666667,"width":0.0265625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"bounds":{"left":0.19648437,"top":0.06527778,"width":0.037890624,"height":0.0125},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"bounds":{"left":0.19648437,"top":0.06666667,"width":0.037890624,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"bounds":{"left":0.2359375,"top":0.06666667,"width":0.068359375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"bounds":{"left":0.3058594,"top":0.06388889,"width":0.021484375,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"bounds":{"left":0.30820313,"top":0.06666667,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"bounds":{"left":0.32890624,"top":0.06666667,"width":0.01015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"bounds":{"left":0.340625,"top":0.06388889,"width":0.1203125,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"bounds":{"left":0.34296876,"top":0.06666667,"width":0.115625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"bounds":{"left":0.4625,"top":0.061805554,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"bounds":{"left":0.78046876,"top":0.060416665,"width":0.00234375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"bounds":{"left":0.7828125,"top":0.060416665,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"bounds":{"left":0.7867187,"top":0.060416665,"width":0.008984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"bounds":{"left":0.796875,"top":0.060416665,"width":0.015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"bounds":{"left":0.8222656,"top":0.05625,"width":0.0546875,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"bounds":{"left":0.8335937,"top":0.060416665,"width":0.03984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"bounds":{"left":0.88007814,"top":0.05625,"width":0.0453125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"bounds":{"left":0.88359374,"top":0.060416665,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"bounds":{"left":0.9007813,"top":0.060416665,"width":0.01484375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"bounds":{"left":0.9285156,"top":0.05625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"bounds":{"left":0.94921875,"top":0.05625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"bounds":{"left":0.96171874,"top":0.05625,"width":0.0203125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"bounds":{"left":0.97148436,"top":0.060416665,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"bounds":{"left":0.9746094,"top":0.060416665,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"bounds":{"left":0.9777344,"top":0.060416665,"width":0.001953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"bounds":{"left":0.11953125,"top":0.09791667,"width":0.080078125,"height":0.020833334},"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"bounds":{"left":0.203125,"top":0.097222224,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"bounds":{"left":0.10664062,"top":0.13055556,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"bounds":{"left":0.10664062,"top":0.13263889,"width":0.016796876,"height":0.045833334},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"bounds":{"left":0.12539062,"top":0.13611111,"width":0.17070313,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"bounds":{"left":0.12851563,"top":0.15833333,"width":0.033203125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"bounds":{"left":0.12851563,"top":0.18055555,"width":0.012890625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"bounds":{"left":0.12539062,"top":0.20277777,"width":0.01953125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"bounds":{"left":0.12851563,"top":0.225,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"bounds":{"left":0.12851563,"top":0.225,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"bounds":{"left":0.12851563,"top":0.24722221,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"bounds":{"left":0.12851563,"top":0.24722221,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"bounds":{"left":0.12539062,"top":0.2701389,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"bounds":{"left":0.12539062,"top":0.2701389,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"bounds":{"left":0.23203126,"top":0.08958333,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"src/eval/eval-ai-activity-type.py","depth":15,"bounds":{"left":0.24453124,"top":0.09236111,"width":0.09296875,"height":0.014583333},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXLink","text":"src/eval/eval-ai-activity-type.py","depth":16,"bounds":{"left":0.24453124,"top":0.09375,"width":0.09296875,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"src/eval/eval-ai-activity-type.py","depth":18,"bounds":{"left":0.24453124,"top":0.09513889,"width":0.09296875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy file name to clipboard","depth":15,"bounds":{"left":0.340625,"top":0.08958333,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 330 additions & 0 deletions","depth":15,"bounds":{"left":0.884375,"top":0.10069445,"width":0.0234375,"height":0.099305555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXCheckBox","text":"Not Viewed","depth":14,"bounds":{"left":0.91914064,"top":0.08958333,"width":0.03125,"height":0.019444445},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Viewed","depth":16,"bounds":{"left":0.93046874,"top":0.094444446,"width":0.01640625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Comment on this file","depth":14,"bounds":{"left":0.95351565,"top":0.08958333,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"More options","depth":14,"bounds":{"left":0.9675781,"top":0.08958333,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Original file line number","depth":17,"bounds":{"left":0.22890624,"top":0.07430556,"width":0.02109375,"height":0.04097222},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Original file line","depth":17,"bounds":{"left":0.25,"top":0.08194444,"width":0.02109375,"height":0.02638889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line number","depth":17,"bounds":{"left":0.27109376,"top":0.07430556,"width":0.02109375,"height":0.04097222},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line change","depth":17,"bounds":{"left":0.2921875,"top":0.07430556,"width":0.01953125,"height":0.04097222},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"@@ -0,0 +1,330 @@","depth":18,"bounds":{"left":0.25546876,"top":0.07638889,"width":0.048046876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1","depth":16,"bounds":{"left":0.6125,"top":0.093055554,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.093055554,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#!/usr/bin/env python3","depth":18,"bounds":{"left":0.63203126,"top":0.093055554,"width":0.06171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"2","depth":16,"bounds":{"left":0.6125,"top":0.10972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.10972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.63203126,"top":0.10972222,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio","depth":18,"bounds":{"left":0.6515625,"top":0.10972222,"width":0.019921875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":16,"bounds":{"left":0.6125,"top":0.12638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.12638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.63203126,"top":0.12638889,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json","depth":18,"bounds":{"left":0.6515625,"top":0.12638889,"width":0.011328125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"4","depth":16,"bounds":{"left":0.6125,"top":0.14305556,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.14305556,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.63203126,"top":0.14305556,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"os","depth":18,"bounds":{"left":0.6515625,"top":0.14305556,"width":0.00546875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"5","depth":16,"bounds":{"left":0.6125,"top":0.15972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.15972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"bounds":{"left":0.63203126,"top":0.15972222,"width":0.0109375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"pathlib","depth":18,"bounds":{"left":0.6460937,"top":0.15972222,"width":0.01953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.6683594,"top":0.15972222,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Path","depth":18,"bounds":{"left":0.68828124,"top":0.15972222,"width":0.011328125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":16,"bounds":{"left":0.6125,"top":0.17638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.17638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"bounds":{"left":0.63203126,"top":0.17638889,"width":0.0109375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"typing","depth":18,"bounds":{"left":0.6460937,"top":0.17638889,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.665625,"top":0.17638889,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Any","depth":18,"bounds":{"left":0.6855469,"top":0.17638889,"width":0.008203125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"bounds":{"left":0.69375,"top":0.17638889,"width":0.005859375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Dict","depth":18,"bounds":{"left":0.6996094,"top":0.17638889,"width":0.011328125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"bounds":{"left":0.7109375,"top":0.17638889,"width":0.00546875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"List","depth":18,"bounds":{"left":0.7164062,"top":0.17638889,"width":0.011328125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"7","depth":16,"bounds":{"left":0.6125,"top":0.19305556,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.19305556,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"8","depth":16,"bounds":{"left":0.6125,"top":0.20972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.20972222,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.63203126,"top":0.20972222,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":18,"bounds":{"left":0.6515625,"top":0.20972222,"width":0.019921875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":16,"bounds":{"left":0.6125,"top":0.22638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.22638889,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"10","depth":16,"bounds":{"left":0.6109375,"top":0.24305555,"width":0.005859375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.24305555,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"bounds":{"left":0.63203126,"top":0.24305555,"width":0.0109375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"bounds":{"left":0.6460937,"top":0.24305555,"width":0.008203125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.6542969,"top":0.24305555,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"bounds":{"left":0.65703124,"top":0.24305555,"width":0.025390625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.68242186,"top":0.24305555,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"bounds":{"left":0.6855469,"top":0.24305555,"width":0.0421875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.7277344,"top":0.24305555,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"bounds":{"left":0.73046875,"top":0.24305555,"width":0.019921875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.75039065,"top":0.24305555,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_request","depth":18,"bounds":{"left":0.753125,"top":0.24305555,"width":0.06757812,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.8234375,"top":0.24305555,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeRequest","depth":18,"bounds":{"left":0.84335935,"top":0.24305555,"width":0.058984376,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"11","depth":16,"bounds":{"left":0.6109375,"top":0.25972223,"width":0.005859375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"bounds":{"left":0.62578124,"top":0.25972223,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"bounds":{"left":0.63203126,"top":0.25972223,"width":0.0109375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"bounds":{"left":0.6460937,"top":0.25972223,"width":0.008203125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.6542969,"top":0.25972223,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"bounds":{"left":0.65703124,"top":0.25972223,"width":0.025390625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.68242186,"top":0.25972223,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"bounds":{"left":0.6855469,"top":0.25972223,"width":0.0421875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.7277344,"top":0.25972223,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"bounds":{"left":0.73046875,"top":0.25972223,"width":0.019921875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"bounds":{"left":0.75039065,"top":0.25972223,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_response","depth":18,"bounds":{"left":0.753125,"top":0.25972223,"width":0.0703125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"bounds":{"left":0.8261719,"top":0.25972223,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeResponse","depth":18,"bounds":{"left":0.8460938,"top":0.25972223,"width":0.062109374,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"12","depth":16,"bounds":{"left":0.6109375,"top":0.27638888,"width":0.005859375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
1909823194848081563
|
6299546254565977410
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import
AiActivityTypeResponse
12...
|
NULL
|
|
8524
|
164
|
0
|
2026-04-14T06:50:10.750263+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149410750_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-f github.com/jiminny/prophet/pull/468/changes#diff-fc8a653dec2cd8e3d632fa5d32e0abd4b22176ce7619cc5752c4efe9e53f083b...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"src/eval/eval-ai-activity-type.py","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXLink","text":"src/eval/eval-ai-activity-type.py","depth":16,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"src/eval/eval-ai-activity-type.py","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy file name to clipboard","depth":15,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 330 additions & 0 deletions","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXCheckBox","text":"Not Viewed","depth":14,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Viewed","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Comment on this file","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"More options","depth":14,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Original file line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Original file line","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line change","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"@@ -0,0 +1,330 @@","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#!/usr/bin/env python3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"2","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"4","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"os","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"5","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"pathlib","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Path","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"typing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Any","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Dict","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"List","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"7","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"8","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"10","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_request","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeRequest","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"11","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_response","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
4153859112112021314
|
6299545704810163522
|
idle
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import...
|
NULL
|
|
8525
|
165
|
1
|
2026-04-14T06:50:11.129931+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149411129_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-f github.com/jiminny/prophet/pull/468/changes#diff-fc8a653dec2cd8e3d632fa5d32e0abd4b22176ce7619cc5752c4efe9e53f083b...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import
AiActivityTypeResponse...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"bounds":{"left":0.10625,"top":0.065972224,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"bounds":{"left":0.10625,"top":0.068055555,"width":0.03515625,"height":0.07083333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"bounds":{"left":0.10625,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"bounds":{"left":0.1328125,"top":0.060416665,"width":0.0140625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"bounds":{"left":0.1546875,"top":0.050694443,"width":0.1046875,"height":0.014583333},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"bounds":{"left":0.1546875,"top":0.052083332,"width":0.1046875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"bounds":{"left":0.2625,"top":0.052083332,"width":0.003515625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"bounds":{"left":0.26601562,"top":0.052083332,"width":0.01015625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"bounds":{"left":0.1515625,"top":0.0625,"width":0.03984375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"bounds":{"left":0.15507813,"top":0.06736111,"width":0.0265625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"bounds":{"left":0.19648437,"top":0.065972224,"width":0.037890624,"height":0.0125},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"bounds":{"left":0.19648437,"top":0.06736111,"width":0.037890624,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"bounds":{"left":0.2359375,"top":0.06736111,"width":0.068359375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"bounds":{"left":0.3058594,"top":0.06458333,"width":0.021484375,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"bounds":{"left":0.30820313,"top":0.06736111,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"bounds":{"left":0.32890624,"top":0.06736111,"width":0.01015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"bounds":{"left":0.340625,"top":0.06458333,"width":0.1203125,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"bounds":{"left":0.34296876,"top":0.06736111,"width":0.115625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"bounds":{"left":0.4625,"top":0.0625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"bounds":{"left":0.78046876,"top":0.06111111,"width":0.00234375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"bounds":{"left":0.7828125,"top":0.06111111,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"bounds":{"left":0.7867187,"top":0.06111111,"width":0.008984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"bounds":{"left":0.796875,"top":0.06111111,"width":0.015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"bounds":{"left":0.8222656,"top":0.056944445,"width":0.0546875,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"bounds":{"left":0.8335937,"top":0.06111111,"width":0.03984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"bounds":{"left":0.88007814,"top":0.056944445,"width":0.0453125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"bounds":{"left":0.88359374,"top":0.06111111,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"bounds":{"left":0.9007813,"top":0.06111111,"width":0.01484375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"bounds":{"left":0.9285156,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"bounds":{"left":0.94921875,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"bounds":{"left":0.96171874,"top":0.056944445,"width":0.0203125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"bounds":{"left":0.97148436,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"bounds":{"left":0.9746094,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"bounds":{"left":0.9777344,"top":0.06111111,"width":0.001953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"bounds":{"left":0.11953125,"top":0.09861111,"width":0.080078125,"height":0.020833334},"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"bounds":{"left":0.203125,"top":0.09791667,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"bounds":{"left":0.10664062,"top":0.13125,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"bounds":{"left":0.10664062,"top":0.13333334,"width":0.016796876,"height":0.045833334},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"bounds":{"left":0.12539062,"top":0.13680555,"width":0.17070313,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"bounds":{"left":0.12851563,"top":0.15902779,"width":0.033203125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"bounds":{"left":0.12851563,"top":0.18125,"width":0.012890625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"bounds":{"left":0.12539062,"top":0.20347223,"width":0.01953125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"bounds":{"left":0.12851563,"top":0.22569445,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"bounds":{"left":0.12851563,"top":0.22569445,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"bounds":{"left":0.12851563,"top":0.24791667,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"bounds":{"left":0.12851563,"top":0.24791667,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"bounds":{"left":0.12539062,"top":0.27083334,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"bounds":{"left":0.12539062,"top":0.27083334,"width":0.023828125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"bounds":{"left":0.23203126,"top":0.090277776,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"src/eval/eval-ai-activity-type.py","depth":15,"bounds":{"left":0.24453124,"top":0.093055554,"width":0.09296875,"height":0.014583333},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXLink","text":"src/eval/eval-ai-activity-type.py","depth":16,"bounds":{"left":0.24453124,"top":0.094444446,"width":0.09296875,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"src/eval/eval-ai-activity-type.py","depth":18,"bounds":{"left":0.24453124,"top":0.09583333,"width":0.09296875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy file name to clipboard","depth":15,"bounds":{"left":0.340625,"top":0.090277776,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 330 additions & 0 deletions","depth":15,"bounds":{"left":0.884375,"top":0.10138889,"width":0.0234375,"height":0.099305555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXCheckBox","text":"Not Viewed","depth":14,"bounds":{"left":0.91914064,"top":0.090277776,"width":0.03125,"height":0.019444445},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Viewed","depth":16,"bounds":{"left":0.93046874,"top":0.09513889,"width":0.01640625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Comment on this file","depth":14,"bounds":{"left":0.95351565,"top":0.090277776,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"More options","depth":14,"bounds":{"left":0.9675781,"top":0.090277776,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Original file line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Original file line","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line number","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Diff line change","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"@@ -0,0 +1,330 @@","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#!/usr/bin/env python3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"2","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"4","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"os","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"5","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"pathlib","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Path","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"typing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Any","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Dict","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"List","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"7","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"8","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"10","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_request","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeRequest","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"11","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"+","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"component","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call_processing","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"schemes","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type_response","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"import","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeResponse","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
1057859308915227409
|
6299545704801774914
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
src/eval/eval-ai-activity-type.py
Copy file name to clipboard
Lines changed: 330 additions & 0 deletions
Not Viewed
Viewed
Comment on this file
More options
Original file line number
Original file line
Diff line number
Diff line change
@@ -0,0 +1,330 @@
1
+
#!/usr/bin/env python3
2
+
import
asyncio
3
+
import
json
4
+
import
os
5
+
from
pathlib
import
Path
6
+
from
typing
import
Any
,
Dict
,
List
7
+
8
+
import
aiohttp
9
+
10
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_request
import
AiActivityTypeRequest
11
+
from
src
.
component
.
call_processing
.
schemes
.
ai_activity_type_response
import
AiActivityTypeResponse...
|
8523
|
|
8529
|
164
|
2
|
2026-04-14T06:50:22.076456+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149422076_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"1","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":".gitignore","depth":19,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":".gitignore","depth":20,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"This page has been optimized for large pull requests.","depth":14,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests.","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Switch to single file mode","depth":13,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Switch to single file mode","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
-2869390503166716359
|
1685608296201596395
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
1
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py
.gitignore
.gitignore
This page has been optimized for large pull requests.
This page has been optimized for large pull requests.
This page has been optimized for large pull requests. Some browser features (like Find on Page or Select All) may not work as expected.
Switch to single file mode
Switch to single file mode
Collapse file...
|
NULL
|
|
8530
|
165
|
4
|
2026-04-14T06:50:22.089834+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149422089_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
2
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Lines changed: 25671 additions & 0 deletions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Conversation (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Conversation","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Commits (9)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Commits","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"9","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Checks (3)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Checks","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"3","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Files changed (203)","depth":16,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Files changed","depth":17,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":18,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Pull Request Toolbar","depth":14,"bounds":{"left":0.10625,"top":0.065972224,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pull Request Toolbar","depth":15,"bounds":{"left":0.10625,"top":0.068055555,"width":0.03515625,"height":0.07083333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Collapse file tree","depth":14,"bounds":{"left":0.10625,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":true},{"role":"AXStaticText","text":"Open","depth":14,"bounds":{"left":0.1328125,"top":0.060416665,"width":0.0140625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Jy 19798 evaluation for ai activity types","depth":14,"bounds":{"left":0.1546875,"top":0.050694443,"width":0.1046875,"height":0.014583333},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":16,"bounds":{"left":0.1546875,"top":0.052083332,"width":0.1046875,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"bounds":{"left":0.2625,"top":0.052083332,"width":0.003515625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"bounds":{"left":0.26601562,"top":0.052083332,"width":0.01015625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"All commits","depth":14,"bounds":{"left":0.1515625,"top":0.0625,"width":0.03984375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"All commits","depth":16,"bounds":{"left":0.15507813,"top":0.06736111,"width":0.0265625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"nikolaybiaivanov","depth":15,"bounds":{"left":0.19648437,"top":0.065972224,"width":0.037890624,"height":0.0125},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"nikolaybiaivanov","depth":16,"bounds":{"left":0.19648437,"top":0.06736111,"width":0.037890624,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"wants to merge 9 commits into","depth":15,"bounds":{"left":0.2359375,"top":0.06736111,"width":0.068359375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"master","depth":15,"bounds":{"left":0.3058594,"top":0.06458333,"width":0.021484375,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"master","depth":16,"bounds":{"left":0.30820313,"top":0.06736111,"width":0.016796876,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"from","depth":16,"bounds":{"left":0.32890624,"top":0.06736111,"width":0.01015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"JY-19798-evaluation-for-ai-activity-types","depth":16,"bounds":{"left":0.340625,"top":0.06458333,"width":0.1203125,"height":0.015277778},"role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"JY-19798-evaluation-for-ai-activity-types","depth":17,"bounds":{"left":0.34296876,"top":0.06736111,"width":0.115625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy head branch name to clipboard","depth":16,"bounds":{"left":0.4625,"top":0.0625,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"2","depth":15,"bounds":{"left":0.7796875,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":15,"bounds":{"left":0.7828125,"top":0.06111111,"width":0.002734375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"203","depth":15,"bounds":{"left":0.7867187,"top":0.06111111,"width":0.008984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"viewed","depth":15,"bounds":{"left":0.796875,"top":0.06111111,"width":0.015625,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":14,"bounds":{"left":0.8222656,"top":0.056944445,"width":0.0546875,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":16,"bounds":{"left":0.8335937,"top":0.06111111,"width":0.03984375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Submit review","depth":14,"bounds":{"left":0.88007814,"top":0.056944445,"width":0.0453125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Submit","depth":16,"bounds":{"left":0.88359374,"top":0.06111111,"width":0.0171875,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"review","depth":16,"bounds":{"left":0.9007813,"top":0.06111111,"width":0.01484375,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Open diff view settings","depth":14,"bounds":{"left":0.9285156,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open overview panel","depth":14,"bounds":{"left":0.94921875,"top":0.056944445,"width":0.0109375,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open comments panel","depth":14,"bounds":{"left":0.96171874,"top":0.056944445,"width":0.0203125,"height":0.019444445},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"(","depth":16,"bounds":{"left":0.97148436,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":16,"bounds":{"left":0.9746094,"top":0.06111111,"width":0.003125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":16,"bounds":{"left":0.9777344,"top":0.06111111,"width":0.001953125,"height":0.010416667},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextField","text":"Filter files…","depth":16,"bounds":{"left":0.11953125,"top":0.09861111,"width":0.080078125,"height":0.020833334},"help_text":"","role_description":"text field","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Filter options","depth":16,"bounds":{"left":0.203125,"top":0.09791667,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"File tree","depth":15,"bounds":{"left":0.10664062,"top":0.13125,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"File tree","depth":16,"bounds":{"left":0.10664062,"top":0.13333334,"width":0.016796876,"height":0.045833334},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable","depth":19,"bounds":{"left":0.12539062,"top":0.13680555,"width":0.17070313,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ground_truth","depth":21,"bounds":{"left":0.12851563,"top":0.15902779,"width":0.033203125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"input","depth":21,"bounds":{"left":0.12851563,"top":0.18125,"width":0.012890625,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"src/eval","depth":19,"bounds":{"left":0.12539062,"top":0.20347223,"width":0.01953125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"endpoints.py","depth":21,"bounds":{"left":0.12851563,"top":0.22569445,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"endpoints.py","depth":22,"bounds":{"left":0.12851563,"top":0.22569445,"width":0.0328125,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"eval-ai-activity-type.py","depth":21,"bounds":{"left":0.12851563,"top":0.24861111,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"eval-ai-activity-type.py","depth":22,"bounds":{"left":0.12851563,"top":0.24861111,"width":0.058984376,"height":0.011805556},"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
5864725084621537271
|
1401881522361609451
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
Lines changed: 25671 additions & 0 deletions
Conversation (3)
Conversation
(
3
)
Commits (9)
Commits
(
9
)
Checks (3)
Checks
(
3
)
Files changed (203)
Files changed
(
203
)
Pull Request Toolbar
Pull Request Toolbar
Collapse file tree
Open
Jy 19798 evaluation for ai activity types
Jy 19798 evaluation for ai activity types
#
468
All commits
All commits
nikolaybiaivanov
nikolaybiaivanov
wants to merge 9 commits into
master
master
from
JY-19798-evaluation-for-ai-activity-types
JY-19798-evaluation-for-ai-activity-types
Copy head branch name to clipboard
2
/
203
viewed
Awaiting approval
Awaiting approval
Submit review
Submit
review
Open diff view settings
Open overview panel
Open comments panel
(
0
)
Filter files…
Filter options
File tree
File tree
eval_output/ai_activity_type/eval_datasets/v2-from-logs-replayable
ground_truth
input
src/eval
endpoints.py
endpoints.py
eval-ai-activity-type.py
eval-ai-activity-type.py...
|
NULL
|
|
8532
|
164
|
3
|
2026-04-14T06:50:29.042406+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149429042_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Agents","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Actions","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Actions","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Projects","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Projects","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Wiki","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Wiki","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Security and quality (20)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Security and quality","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"20","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Insights","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Insights","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Settings","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Settings","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Important update","depth":10,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Important update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Review this update","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Review this update","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and manage your preferences in your","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"GitHub account settings","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"GitHub account settings","depth":11,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Dismiss banner","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Jy 19798 evaluation for ai activity types #468 Edit title","depth":13,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"#","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"468","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Edit title","depth":14,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Preview","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Preview","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Awaiting approval","depth":13,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Awaiting approval","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXMenuButton","text":"Code","depth":13,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Code","depth":15,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Open","depth":13,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
5108698256320021947
|
-2054631194396840785
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Open Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents
Agents
Actions
Actions
Projects
Projects
Wiki
Wiki
Security and quality (20)
Security and quality
(
20
)
Insights
Insights
Settings
Settings
Important update
Important update
On April 24 we'll start using GitHub Copilot interaction data for AI model training unless you opt out.
Review this update
Review this update
and manage your preferences in your
GitHub account settings
GitHub account settings
.
Dismiss banner
Jy 19798 evaluation for ai activity types #468 Edit title
Jy 19798 evaluation for ai activity types
#
468
Edit title
Preview
Preview
Awaiting approval
Awaiting approval
Code
Code
Open...
|
8529
|
|
8542
|
165
|
11
|
2026-04-14T06:51:07.263345+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149467263_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Analyzing Implementation
Gemini said
Gemini said
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools
Open mode picker
Pro
Stop response
Your Jiminny chats aren’t used to improve our models. Gemini is AI and can make mistakes, including about people.
Your privacy & Gemini Opens in a new window
Your privacy & Gemini
Opens in a new window
Summarize page
Summarize page
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.0,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.0,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.0,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.0,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.0,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.024305556,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.024305556,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.04375,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.04236111,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.15972222,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.15972222,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":false,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.16805555,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.1701389,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.17083333,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.22986111,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Analyzing Implementation","depth":26,"bounds":{"left":0.11953125,"top":0.23680556,"width":0.06679688,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.27291667,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.27430555,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextArea","text":"Enter a prompt for Gemini\nencrypted","depth":20,"bounds":{"left":0.109375,"top":0.8354167,"width":0.125,"height":0.016666668},"value":"Enter a prompt for Gemini\nencrypted","help_text":"","role_description":"text entry area","subrole":"AXUnknown","is_enabled":true,"is_focused":true,"is_selected":false},{"role":"AXStaticText","text":"Enter a prompt for Gemini","depth":21,"bounds":{"left":0.1171875,"top":0.8354167,"width":0.08203125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"encrypted","depth":21,"bounds":{"left":0.10820313,"top":0.8354167,"width":0.0078125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open upload file menu","depth":20,"bounds":{"left":0.1046875,"top":0.86527777,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Tools","depth":18,"bounds":{"left":0.1234375,"top":0.86527777,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open mode picker","depth":20,"bounds":{"left":0.190625,"top":0.8645833,"width":0.03046875,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pro","depth":23,"bounds":{"left":0.196875,"top":0.87222224,"width":0.00859375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Stop response","depth":19,"bounds":{"left":0.22265625,"top":0.86388886,"width":0.01640625,"height":0.029166667},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Your Jiminny chats aren’t used to improve our models. Gemini is AI and can make mistakes, including about people.","depth":17,"bounds":{"left":0.10039063,"top":0.9097222,"width":0.14296874,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Your privacy & Gemini Opens in a new window","depth":17,"bounds":{"left":0.1484375,"top":0.93194443,"width":0.046875,"height":0.011111111},"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Your privacy & Gemini","depth":18,"bounds":{"left":0.1484375,"top":0.93194443,"width":0.046875,"height":0.011111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Opens in a new window","depth":19,"bounds":{"left":0.09335937,"top":0.93125,"width":0.05078125,"height":0.011111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Summarize page","depth":7,"bounds":{"left":0.1,"top":0.96319443,"width":0.06289063,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Summarize page","depth":9,"bounds":{"left":0.10664062,"top":0.9673611,"width":0.049609374,"height":0.013888889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Skip to content","depth":6,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Skip to content","depth":7,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open menu","depth":10,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Homepage (g then d)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"jiminny","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"jiminny","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"prophet","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"prophet","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Search or jump to…","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Type","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"/","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to search","depth":12,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Chat with Copilot","depth":10,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXMenuButton","text":"Open Copilot…","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXMenuButton","text":"Create new...","depth":9,"help_text":"","role_description":"menu button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXLink","text":"Issues","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Pull requests","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"Repositories","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXLink","text":"You have unread notifications(g then n)","depth":9,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open user navigation menu","depth":9,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Repository navigation","depth":9,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Repository navigation","depth":10,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Code","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Code","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Issues","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Issues","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Pull requests (6)","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Pull requests","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"6","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":")","depth":14,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXLink","text":"Agents","depth":12,"help_text":"","role_description":"link","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
6142532471012435964
|
8731758629248348295
|
idle
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Analyzing Implementation
Gemini said
Gemini said
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools
Open mode picker
Pro
Stop response
Your Jiminny chats aren’t used to improve our models. Gemini is AI and can make mistakes, including about people.
Your privacy & Gemini Opens in a new window
Your privacy & Gemini
Opens in a new window
Summarize page
Summarize page
Skip to content
Skip to content
Open menu
Homepage (g then d)
jiminny
jiminny
prophet
prophet
Search or jump to…
Type
/
to search
Chat with Copilot
Open Copilot…
Create new...
Issues
Pull requests
Repositories
You have unread notifications(g then n)
Open user navigation menu
Repository navigation
Repository navigation
Code
Code
Issues
Issues
Pull requests (6)
Pull requests
(
6
)
Agents...
|
8539
|
|
8543
|
164
|
8
|
2026-04-14T06:51:18.848279+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149478848_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Show thinking
Gemini said
Gemini said
Yes,
there are several issues and code smells in this Pull Request,
ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues,
categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
,
the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one.
This means the script waits for one HTTP request to finish before starting the next,
completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":false,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes,","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"there are several issues and code smells in this Pull Request,","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues,","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This means the script waits for one HTTP request to finish before starting the next,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
4108366247784106262
|
8731758625087598999
|
idle
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Show thinking
Gemini said
Gemini said
Yes,
there are several issues and code smells in this Pull Request,
ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues,
categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
,
the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one.
This means the script waits for one HTTP request to finish before starting the next,
completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use...
|
NULL
|
|
8544
|
165
|
12
|
2026-04-14T06:51:22.047656+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149482047_m2.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Show thinking
Gemini said
Gemini said
Yes,
there are several issues and code smells in this Pull Request,
ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues,
categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
,
the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one.
This means the script waits for one HTTP request to finish before starting the next,
completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions.
In Python,
this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
,
or run the blocking I/O operations in a separate thread using
asyncio.to_thread()...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.11736111,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.14097223,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.0,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.0,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.0,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.0,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.0,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.0,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.024305556,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.024305556,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.04375,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.04236111,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.15972222,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.15972222,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":false,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.16805555,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.1701389,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.17083333,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.22986111,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.23680556,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.27291667,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.27430555,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes,","depth":24,"bounds":{"left":0.10390625,"top":0.28125,"width":0.011328125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"there are several issues and code smells in this Pull Request,","depth":24,"bounds":{"left":0.10390625,"top":0.28125,"width":0.12890625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.29930556,"width":0.13242188,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues,","depth":24,"bounds":{"left":0.10390625,"top":0.34652779,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"categorized by impact:","depth":24,"bounds":{"left":0.123046875,"top":0.36458334,"width":0.06679688,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.4,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.40138888,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.42430556,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.44236112,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.46180555,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.17460938,"top":0.46041667,"width":0.001953125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.46041667,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.4798611,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.47847223,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.4798611,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.47847223,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.49791667,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one.","depth":26,"bounds":{"left":0.11757813,"top":0.4965278,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This means the script waits for one HTTP request to finish before starting the next,","depth":26,"bounds":{"left":0.11757813,"top":0.51458335,"width":0.11289062,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.55069447,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.5701389,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.56875,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.61041665,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.61041665,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.6298611,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.6284722,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.66597223,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.6645833,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.72430557,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.7423611,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.7798611,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.77847224,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.7798611,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.77847224,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.79791665,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.7965278,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.79791665,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions.","depth":26,"bounds":{"left":0.11757813,"top":0.7965278,"width":0.11796875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"In Python,","depth":26,"bounds":{"left":0.14570312,"top":0.81458336,"width":0.030078124,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.81458336,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.87430555,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.87430555,"width":0.09335937,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":0.89375,"width":0.0265625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":28,"bounds":{"left":0.19140625,"top":0.8923611,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":0.8923611,"width":0.10585938,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":0.9298611,"width":0.0625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
2043951250885907759
|
9020551951192732151
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Show more options
Show thinking
Gemini said
Gemini said
Yes,
there are several issues and code smells in this Pull Request,
ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues,
categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
,
the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one.
This means the script waits for one HTTP request to finish before starting the next,
completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions.
In Python,
this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
,
or run the blocking I/O operations in a separate thread using
asyncio.to_thread()...
|
NULL
|
|
8546
|
164
|
9
|
2026-04-14T06:51:26.123519+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149486123_m1.jpg...
|
Firefox
|
Jy 19798 evaluation for ai activity types by nikol Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet — Work...
|
1
|
github.com/jiminny/prophet/pull/468/changes#diff-d github.com/jiminny/prophet/pull/468/changes#diff-d2f9fea9bc428b119ae26a66453e8eab7dd0782d4a4d93075ca5ce05903d77e8...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0,"top":0.84555554,"width":0.0027777778,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"bounds":{"left":0.0,"top":0.97,"width":0.018055556,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"bounds":{"left":0.0,"top":0.9988889,"width":0.060416665,"height":0.0011110902},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"bounds":{"left":0.0,"top":1.0,"width":0.0055555557,"height":-0.0011111498},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
4082059317752290689
|
6426478565827391895
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using...
|
8543
|
|
8552
|
164
|
12
|
2026-04-14T06:51:35.430819+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149495430_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:51
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:51","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.020833334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.044444446,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.07986111,"top":0.90444446,"width":0.08125,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.07986111,"top":0.93777776,"width":0.08125,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
6522371285992393864
|
6113841948802480952
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:51
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
NULL
|
|
8553
|
165
|
17
|
2026-04-14T06:51:35.446804+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149495446_m2.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:51
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.068750024},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.06944442},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.7296875,"top":1.0,"width":0.023046875,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.7453125,"top":1.0,"width":0.002734375,"height":-0.068750024},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.7558594,"top":1.0,"width":0.0140625,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.75742185,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.7746094,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.7734375,"top":1.0,"width":0.01328125,"height":-0.0625},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:51","depth":10,"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.23554687,"top":1.0,"width":0.000390625,"height":-0.050694466},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.2957031,"top":1.0,"width":0.12148438,"height":-0.02569449},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
6522371285992393864
|
6113841948802480952
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:51
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8550
|
|
8570
|
164
|
22
|
2026-04-14T06:53:04.602359+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149584602_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:53","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022222223,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.046527777,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08125,"top":0.90444446,"width":0.08194444,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08125,"top":0.93777776,"width":0.08194444,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
4595186536147244130
|
6113912317546658616
|
visual_change
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8569
|
|
8571
|
165
|
25
|
2026-04-14T06:53:28.788572+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149608788_m2.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.068750024},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.06944442},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.7296875,"top":1.0,"width":0.023046875,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.7453125,"top":1.0,"width":0.002734375,"height":-0.068750024},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.7558594,"top":1.0,"width":0.0140625,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.75742185,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.7746094,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.7734375,"top":1.0,"width":0.01328125,"height":-0.0625},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:53","depth":10,"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.23554687,"top":1.0,"width":0.000390625,"height":-0.050694466},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.2957031,"top":1.0,"width":0.12148438,"height":-0.02569449},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
4595186536147244130
|
6113912317546658616
|
idle
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8568
|
|
8572
|
164
|
23
|
2026-04-14T06:53:37.920446+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149617920_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:53","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022222223,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.046527777,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08125,"top":0.90444446,"width":0.08194444,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08125,"top":0.93777776,"width":0.08194444,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
4595186536147244130
|
6113912317546658616
|
visual_change
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
NULL
|
|
8574
|
164
|
25
|
2026-04-14T06:53:50.015198+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149630015_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:53","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022222223,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.046527777,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08125,"top":0.90444446,"width":0.08194444,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08125,"top":0.93777776,"width":0.08194444,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
4595186536147244130
|
6113912317546658616
|
visual_change
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
NULL
|
|
8575
|
165
|
26
|
2026-04-14T06:53:59.033351+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149639033_m2.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.068750024},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.06944442},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.7296875,"top":1.0,"width":0.023046875,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.7453125,"top":1.0,"width":0.002734375,"height":-0.068750024},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.7558594,"top":1.0,"width":0.0140625,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.75742185,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.7746094,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.7734375,"top":1.0,"width":0.01328125,"height":-0.0625},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:53","depth":10,"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.23554687,"top":1.0,"width":0.000390625,"height":-0.050694466},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.2957031,"top":1.0,"width":0.12148438,"height":-0.02569449},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
4595186536147244130
|
6113912317546658616
|
idle
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:53
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
NULL
|
|
8576
|
164
|
26
|
2026-04-14T06:54:02.157795+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149642157_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:54
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:54","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022916667,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.047222223,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08194444,"top":0.90444446,"width":0.08194444,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08194444,"top":0.93777776,"width":0.08194444,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
-6448776714852722509
|
6113841948802480956
|
visual_change
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:54
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8574
|
|
8584
|
164
|
32
|
2026-04-14T06:54:44.462543+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149684462_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:54
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:54","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022916667,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.047222223,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08194444,"top":0.90444446,"width":0.08194444,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08194444,"top":0.93777776,"width":0.08194444,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
-6448776714852722509
|
6113841948802480956
|
visual_change
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:54
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8583
|
|
8591
|
167
|
0
|
2026-04-14T06:55:06.977785+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149706977_m2.jpg...
|
Firefox
|
Work — Mozilla Firefox
|
1
|
app.staging.jiminny.com
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
New Tab
New Tab
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters....
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.06458333,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.10208333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.11180556,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.13055556,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.14027777,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"New Tab","depth":4,"bounds":{"left":0.0,"top":0.15902779,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"New Tab","depth":5,"bounds":{"left":0.015625,"top":0.16875,"width":0.017578125,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.16527778,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.18888889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.06736111,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.06736111,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.10972222,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.10972222,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.10972222,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.10972222,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.14791666,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.15,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.004166667,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0055555557,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.028472222,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.029861111,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.047916666,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.06458333,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.065972224,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.06458333,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.084027775,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.08263889,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.10625,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.10625,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.16597222,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.16597222,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.18541667,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.18402778,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.3013889,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.3013889,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.30972221,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.31180555,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.3125,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.37638888,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.3715278,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.3784722,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.41458333,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.41597223,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.42291668,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.48819444,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5416667,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.54305553,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5659722,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.58402777,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.60347223,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.6020833,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6215278,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6201389,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6215278,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6201389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.63958335,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.63819444,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.7118056,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.7104167,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.75208336,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.75208336,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.77152777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.77013886,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.8076389,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.80625,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.8659722,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8840278,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.9215278,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.9201389,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.9215278,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.9201389,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.93958336,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.93819445,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.93958336,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.93819445,"width":0.11875,"height":0.061805546},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.009765625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.09335937,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.035416722},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.034027815},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.07152772},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.07013893},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"{es_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"gt_{req_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":". If","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"req_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"originate from a database and ever contain characters like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"../","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", it could result in files being read from or written to unintended directories.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
4930158969472011583
|
8732321575041020375
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
New Tab
New Tab
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters....
|
NULL
|
|
8593
|
166
|
0
|
2026-04-14T06:55:10.956179+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149710956_m1.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
5074366443921049573
|
8731758625087598999
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:...
|
8588
|
|
8604
|
167
|
10
|
2026-04-14T06:55:37.342012+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149737342_m2.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.14583333,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.16944444,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"{es_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"gt_{req_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":". If","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
8505520146988350698
|
6426478565827391895
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If...
|
NULL
|
|
8607
|
166
|
4
|
2026-04-14T06:55:41.411577+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149741411_m1.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0013888889,"top":0.08111111,"width":0.0027777778,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0,"top":1.0,"width":0.0027777778,"height":-0.04111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
7596023454092539860
|
6426478565827326231
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to...
|
8605
|
|
8609
|
167
|
13
|
2026-04-14T06:55:43.422806+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149743422_m2.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.14583333,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.16944444,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"{es_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"gt_{req_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":". If","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"req_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"originate from a database and ever contain characters like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"../","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", it could result in files being read from or written to unintended directories.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Would you like me to rewrite the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"method to show how to fix the sequential async loop and blocking file I/O?","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextArea","text":"Enter a prompt for Gemini\nencrypted","depth":20,"bounds":{"left":0.109375,"top":0.8354167,"width":0.125,"height":0.016666668},"value":"Enter a prompt for Gemini\nencrypted","help_text":"","role_description":"text entry area","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Enter a prompt for Gemini","depth":21,"bounds":{"left":0.1171875,"top":0.8354167,"width":0.08203125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"encrypted","depth":21,"bounds":{"left":0.10820313,"top":0.8354167,"width":0.0078125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open upload file menu","depth":20,"bounds":{"left":0.1046875,"top":0.86527777,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false}]...
|
-6264578760586490979
|
8732321574906802243
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu...
|
8608
|
|
8611
|
166
|
5
|
2026-04-14T06:55:55.704139+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149755704_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:55
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"bounds":{"left":0.87222224,"top":0.36666667,"width":0.077083334,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"bounds":{"left":0.7430556,"top":0.6188889,"width":0.09097222,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"bounds":{"left":0.87222224,"top":0.6188889,"width":0.075,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"bounds":{"left":0.7423611,"top":0.8711111,"width":0.06944445,"height":0.022222223},"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"bounds":{"left":0.9618056,"top":0.86777776,"width":0.018055556,"height":0.028888889},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:55","depth":10,"bounds":{"left":0.020833334,"top":0.93777776,"width":0.022916667,"height":0.022222223},"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"bounds":{"left":0.046527777,"top":0.93777776,"width":0.018055556,"height":0.022222223},"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"bounds":{"left":0.08194444,"top":0.90444446,"width":0.08125,"height":0.08888889},"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"bounds":{"left":0.08194444,"top":0.93777776,"width":0.08125,"height":0.022222223},"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"bounds":{"left":0.30416667,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"bounds":{"left":0.33194444,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"bounds":{"left":0.37083334,"top":0.9222222,"width":0.06111111,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"bounds":{"left":0.3986111,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"bounds":{"left":0.4375,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"bounds":{"left":0.48194444,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"bounds":{"left":0.5263889,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"bounds":{"left":0.5708333,"top":0.9222222,"width":0.03888889,"height":0.053333335},"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"bounds":{"left":0.61527777,"top":0.9222222,"width":0.025,"height":0.053333335},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"bounds":{"left":0.6458333,"top":0.9222222,"width":0.05,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"bounds":{"left":0.8875,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"bounds":{"left":0.92083335,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"bounds":{"left":0.95416665,"top":0.9222222,"width":0.033333335,"height":0.053333335},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.004166667,"top":0.08111111,"width":0.00069444446,"height":0.0011111111},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.11111111,"top":0.04111111,"width":0.21597221,"height":0.033333335},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
6194137157482527795
|
6113912317546658616
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:55
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
NULL
|
|
8612
|
167
|
15
|
2026-04-14T06:55:55.668931+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149755668_m2.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:55
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.068750024},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.06944442},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.7296875,"top":1.0,"width":0.023046875,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.7453125,"top":1.0,"width":0.002734375,"height":-0.068750024},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.7558594,"top":1.0,"width":0.0140625,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.75742185,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.7746094,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.7734375,"top":1.0,"width":0.01328125,"height":-0.0625},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Yankov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Steliyan Georgiev","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Nikolay Ivanov","depth":17,"role_description":"text"},{"role":"AXStaticText","text":"Lukas Kovalik","depth":17,"role_description":"text"},{"role":"AXButton","text":"Others might see more of your background. Click to view your full video.","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"9:55","depth":10,"role_description":"text"},{"role":"AXStaticText","text":"AM","depth":10,"role_description":"text"},{"role":"AXHeading","text":"Daily - Platform","depth":10,"role_description":"heading"},{"role":"AXStaticText","text":"Daily - Platform","depth":13,"role_description":"text"},{"role":"AXPopUpButton","text":"Audio settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on microphone","depth":14,"role_description":"button","is_enabled":true,"is_focused":true,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"Video settings","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn off camera","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Nikolay Yankov is presenting","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Send a reaction","depth":12,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Turn on captions","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXCheckBox","text":"Raise hand (ctrl + ⌘ + h)","depth":11,"role_description":"toggle button","subrole":"AXToggleButton","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXPopUpButton","text":"More options","depth":14,"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Leave call","depth":11,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting details","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Chat with everyone","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Meeting tools","depth":12,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Your microphone is off.","depth":4,"bounds":{"left":0.23554687,"top":1.0,"width":0.000390625,"height":-0.050694466},"role_description":"text"},{"role":"AXTextArea","text":"meet.google.com / Daily - Platform","depth":5,"bounds":{"left":0.2957031,"top":1.0,"width":0.12148438,"height":-0.02569449},"automation_id":"navigationBarAssistantBarTextField","value":"meet.google.com / Daily - Platform","role_description":"text entry area","is_focused":false}]...
|
6194137157482527795
|
6113912317546658616
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova
Nikolay Yankov
Steliyan Georgiev
Nikolay Ivanov
Lukas Kovalik
Others might see more of your background. Click to view your full video.
9:55
AM
Daily - Platform
Daily - Platform
Audio settings
Turn on microphone
Video settings
Turn off camera
Nikolay Yankov is presenting
Send a reaction
Turn on captions
Raise hand (ctrl + ⌘ + h)
More options
Leave call
Meeting details
Chat with everyone
Meeting tools
Your microphone is off.
meet.google.com / Daily - Platform...
|
8610
|
|
8615
|
166
|
7
|
2026-04-14T06:56:10.121660+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149770121_m1.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
-4024085138441079458
|
6160734901706043051
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
New Tab
Customize sidebar...
|
8613
|
|
8620
|
167
|
20
|
2026-04-14T06:56:15.044426+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149775044_m2.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.19791667,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"{es_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"gt_{req_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":". If","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"req_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"originate from a database and ever contain characters like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"../","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", it could result in files being read from or written to unintended directories.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Would you like me to rewrite the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"method to show how to fix the sequential async loop and blocking file I/O?","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextArea","text":"Enter a prompt for Gemini\nencrypted","depth":20,"bounds":{"left":0.109375,"top":0.8354167,"width":0.125,"height":0.016666668},"value":"Enter a prompt for Gemini\nencrypted","help_text":"","role_description":"text entry area","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Enter a prompt for Gemini","depth":21,"bounds":{"left":0.1171875,"top":0.8354167,"width":0.08203125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"encrypted","depth":21,"bounds":{"left":0.10820313,"top":0.8354167,"width":0.0078125,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open upload file menu","depth":20,"bounds":{"left":0.1046875,"top":0.86527777,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Tools","depth":18,"bounds":{"left":0.1234375,"top":0.86527777,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false}]...
|
-1804690491028682382
|
8732321575042068887
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools...
|
NULL
|
|
8623
|
167
|
21
|
2026-04-14T06:56:45.391003+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149805391_m2.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.19791667,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
479746445631130631
|
5850017813519773975
|
idle
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):...
|
8620
|
|
8624
|
166
|
11
|
2026-04-14T06:56:55.430786+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149815430_m1.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
2367160298329268996
|
8732321579331784839
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM...
|
8622
|
|
8625
|
167
|
22
|
2026-04-14T06:56:55.395589+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149815395_m2.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.19791667,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
5035648433098391144
|
5850017813519708295
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:...
|
NULL
|
|
8629
|
167
|
25
|
2026-04-14T06:57:00.689376+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149820689_m2.jpg...
|
Firefox
|
Jiminny Mail — Work
|
1
|
mail.google.com/mail/u/0/#inbox
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Jiminny Mail
Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.025,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.20277777,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
6131584864387625534
|
6138248189671420039
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Jiminny Mail
Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS...
|
8628
|
|
8633
|
166
|
13
|
2026-04-14T06:57:23.760141+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149843760_m1.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.045833334,"top":0.11,"width":0.125,"height":0.022222223},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.045833334,"top":0.11111111,"width":0.125,"height":0.02},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.8826389,"top":0.09888889,"width":0.04097222,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.91041666,"top":0.11,"width":0.0048611113,"height":0.016666668},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.9291667,"top":0.09888889,"width":0.025,"height":0.04},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.93194443,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.9625,"top":0.11,"width":0.022222223,"height":0.016666668},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.9604167,"top":0.1,"width":0.023611112,"height":0.037777778},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"bounds":{"left":0.6201389,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"bounds":{"left":0.65347224,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"bounds":{"left":0.68680555,"top":0.7922222,"width":0.027777778,"height":0.044444446},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"bounds":{"left":0.7430556,"top":0.36666667,"width":0.08888889,"height":0.022222223},"role_description":"text"}]...
|
6319403837425361321
|
8077623044035967784
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova...
|
8627
|
|
8634
|
167
|
29
|
2026-04-14T06:57:23.760115+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149843760_m2.jpg...
|
Dia
|
Work: Meet - Daily - Pla…
|
1
|
NULL
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova...
|
[{"role":"AXHeading","text" [{"role":"AXHeading","text":"Nikolay Yankov (Presenting)","depth":13,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.068750024},"role_description":"heading"},{"role":"AXStaticText","text":"Nikolay Yankov (Presenting)","depth":14,"bounds":{"left":0.2589844,"top":1.0,"width":0.0703125,"height":-0.06944442},"role_description":"text"},{"role":"AXPopUpButton","text":"People","depth":15,"bounds":{"left":0.7296875,"top":1.0,"width":0.023046875,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"6","depth":18,"bounds":{"left":0.7453125,"top":1.0,"width":0.002734375,"height":-0.068750024},"role_description":"text"},{"role":"AXPopUpButton","text":"Take notes with Gemini","depth":15,"bounds":{"left":0.7558594,"top":1.0,"width":0.0140625,"height":-0.061805606},"role_description":"pop up button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Take notes with Gemini","depth":19,"bounds":{"left":0.75742185,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXStaticText","text":"Gemini","depth":19,"bounds":{"left":0.7746094,"top":1.0,"width":0.0125,"height":-0.068750024},"role_description":"text"},{"role":"AXButton","text":"Gemini","depth":18,"bounds":{"left":0.7734375,"top":1.0,"width":0.01328125,"height":-0.0625},"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Zoom in","depth":14,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open in new window","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Enter Full Screen","depth":15,"role_description":"button","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXStaticText","text":"Stefka Stoyanova","depth":17,"role_description":"text"}]...
|
6319403837425361321
|
8077623044035967784
|
click
|
accessibility
|
NULL
|
Nikolay Yankov (Presenting)
Nikolay Yankov (Presen Nikolay Yankov (Presenting)
Nikolay Yankov (Presenting)
People
6
Take notes with Gemini
Take notes with Gemini
Gemini
Gemini
Zoom in
Open in new window
Enter Full Screen
Stefka Stoyanova...
|
8632
|
|
8638
|
166
|
16
|
2026-04-14T06:57:37.300197+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149857300_m1.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
-5107577173204566284
|
1545677505700740271
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab...
|
NULL
|
|
8639
|
167
|
31
|
2026-04-14T06:57:37.300185+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149857300_m2.jpg...
|
Firefox
|
Jiminny — Work
|
1
|
app.staging.jiminny.com/ai-reports
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.14583333,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.11484375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
-5107577173204566284
|
1545677505700740271
|
click
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Close tab
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab...
|
8636
|
|
8642
|
166
|
18
|
2026-04-14T06:57:44.468996+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149864468_m1.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0013888889,"top":0.08111111,"width":0.0027777778,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0,"top":1.0,"width":0.0027777778,"height":-0.04111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
-2553411110187335245
|
6426478565827391767
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to...
|
8640
|
|
8643
|
167
|
33
|
2026-04-14T06:57:44.978037+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149864978_m2.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.11484375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
8448672044444230511
|
6714708941980086551
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError...
|
8641
|
|
8647
|
166
|
20
|
2026-04-14T06:57:59.610437+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149879610_m1.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools
Open mode picker...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0013888889,"top":0.08111111,"width":0.0027777778,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0,"top":1.0,"width":0.0027777778,"height":-0.04111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"{es_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"f\"gt_{req_id}.json\"","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":". If","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"req_id","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"originate from a database and ever contain characters like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"../","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", it could result in files being read from or written to unintended directories.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Would you like me to rewrite the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"method to show how to fix the sequential async loop and blocking file I/O?","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXTextArea","text":"Enter a prompt for Gemini\nencrypted","depth":20,"value":"Enter a prompt for Gemini\nencrypted","help_text":"","role_description":"text entry area","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Enter a prompt for Gemini","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"encrypted","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Open upload file menu","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Tools","depth":18,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Open mode picker","depth":20,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false}]...
|
-8457584461451125948
|
6426478565828374791
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using
f"{es_id}.json"
and
f"gt_{req_id}.json"
. If
es_id
or
req_id
originate from a database and ever contain characters like
../
, it could result in files being read from or written to unintended directories.
Fix:
Sanitize the IDs or strictly enforce that they only contain alphanumeric characters.
Would you like me to rewrite the
_collect_from_es
method to show how to fix the sequential async loop and blocking file I/O?
Enter a prompt for Gemini
encrypted
Enter a prompt for Gemini
encrypted
Open upload file menu
Tools
Open mode picker...
|
8644
|
|
8648
|
167
|
36
|
2026-04-14T06:58:21.595008+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149901595_m2.jpg...
|
Firefox
|
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf — Work...
|
1
|
app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d0 app.staging.jiminny.com/ai-reports/pdf/e2f4890a-d097-495b-8dfd-d4c25a5e8843...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.17430556,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.11484375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
7070922051797381983
|
6426478565827391895
|
idle
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Close tab
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):...
|
NULL
|
|
8651
|
167
|
38
|
2026-04-14T06:58:27.484149+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149907484_m2.jpg...
|
Firefox
|
Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminn Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail — Work...
|
1
|
mail.google.com/mail/u/0/#inbox
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.11484375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.20277777,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
8537606261635373144
|
6138248189672468887
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()...
|
NULL
|
|
8652
|
166
|
22
|
2026-04-14T06:58:29.817167+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149909817_m1.jpg...
|
Firefox
|
Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminn Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail — Work...
|
1
|
mail.google.com/mail/u/0/#inbox
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Inbox (1,540) - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
7018639353123991036
|
6714708941975892375
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Inbox (1,540) - [EMAIL] - Jiminny Mail
Inbox (1,540) - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (...
|
8649
|
|
8667
|
166
|
30
|
2026-04-14T06:59:06.025467+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149946025_m1.jpg...
|
Firefox
|
Your Ask Jiminny test report report is ready - luk Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail — Work...
|
1
|
mail.google.com/mail/u/0/#inbox/FMfcgzQgLPHqhfTqvc mail.google.com/mail/u/0/#inbox/FMfcgzQgLPHqhfTqvcRjxxxrPDNRGZTl...
|
monitor_1
|
NULL
|
NULL
|
NULL
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using...
|
[{"role":"AXRadioButton","text [{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0013888889,"top":0.08111111,"width":0.0027777778,"height":0.023333333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.0,"top":1.0,"width":0.0027777778,"height":-0.04111111},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"just","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to check its length. Since","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"defaults to","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"if min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Potential Path Traversal Risk (Lines 110, 113, 251):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script writes and reads files directly using","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
-2892713408067612165
|
6426478565828440471
|
visual_change
|
accessibility
|
NULL
|
JY-20543 add AJ reports User pilot tracking by Lak JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string
just
to check its length. Since
min_size
defaults to
0
, the
if min_size
short-circuit saves it for now, but if anyone sets a minimum size, this will become an unnecessary CPU bottleneck on large datasets.
Potential Path Traversal Risk (Lines 110, 113, 251):
The script writes and reads files directly using...
|
8666
|
|
8668
|
167
|
46
|
2026-04-14T06:59:12.976303+00:00
|
/Users/lukas/.screenpipe/data/data/2026-04-14/1776 /Users/lukas/.screenpipe/data/data/2026-04-14/1776149952976_m2.jpg...
|
Firefox
|
Your Ask Jiminny test report report is ready - luk Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail — Work...
|
1
|
mail.google.com/mail/u/0/#inbox/FMfcgzQgLPHqhfTqvc mail.google.com/mail/u/0/#inbox/FMfcgzQgLPHqhfTqvcRjxxxrPDNRGZTl...
|
monitor_2
|
NULL
|
NULL
|
NULL
|
NULL
|
Platform Sprint 1 Q2 - Platform Team - Scrum Board Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
jiminny.atlassian.net
JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Close tab
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string...
|
[{"role":"AXStaticText","text& [{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.09804688,"top":0.08958333,"width":0.09648438,"height":0.019444445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"jiminny.atlassian.net","depth":4,"bounds":{"left":0.09804688,"top":0.10902778,"width":0.04140625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app","depth":4,"bounds":{"left":0.00234375,"top":0.045138888,"width":0.0890625,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":4,"bounds":{"left":0.0,"top":0.08263889,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira","depth":5,"bounds":{"left":0.015625,"top":0.09236111,"width":0.11796875,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.08888889,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXRadioButton","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":4,"bounds":{"left":0.0,"top":0.11111111,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet","depth":5,"bounds":{"left":0.015625,"top":0.12083333,"width":0.1984375,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Jiminny","depth":4,"bounds":{"left":0.0,"top":0.13958333,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Jiminny","depth":5,"bounds":{"left":0.015625,"top":0.14930555,"width":0.015625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":4,"bounds":{"left":0.0,"top":0.16805555,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf","depth":5,"bounds":{"left":0.015625,"top":0.17777778,"width":0.1640625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXRadioButton","text":"Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail","depth":4,"bounds":{"left":0.0,"top":0.19652778,"width":0.09375,"height":0.028472222},"help_text":"","role_description":"tab","subrole":"AXTabButton","is_enabled":true,"is_focused":false,"is_selected":true},{"role":"AXStaticText","text":"Your Ask Jiminny test report report is ready - lukas.kovalik@jiminny.com - Jiminny Mail","depth":5,"bounds":{"left":0.015625,"top":0.20625,"width":0.1765625,"height":0.009722223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Close tab","depth":5,"bounds":{"left":0.07890625,"top":0.20277777,"width":0.009375,"height":0.016666668},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Tab","depth":4,"bounds":{"left":0.003125,"top":0.22638889,"width":0.08710937,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Customize sidebar","depth":6,"bounds":{"left":0.003125,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Close Google Gemini (⌃X)","depth":6,"bounds":{"left":0.01640625,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Tabs from other devices","depth":6,"bounds":{"left":0.029296875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open history (⇧⌘H)","depth":6,"bounds":{"left":0.0421875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXCheckBox","text":"Open bookmarks (⌘B)","depth":6,"bounds":{"left":0.05546875,"top":0.97430557,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"toggle button","subrole":"AXToggle","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"AI Chat settings","depth":7,"bounds":{"left":0.2171875,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Close","depth":7,"bounds":{"left":0.23125,"top":0.047916666,"width":0.0125,"height":0.022222223},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"WORK, Google Account: lukas.kovalik@jiminny.com","depth":12,"bounds":{"left":0.228125,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXButton","text":"Main menu","depth":12,"bounds":{"left":0.0984375,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"New Chat","depth":12,"bounds":{"left":0.1953125,"top":0.090277776,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Open menu for conversation actions.","depth":12,"bounds":{"left":0.2109375,"top":0.090277776,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false,"is_expanded":false},{"role":"AXHeading","text":"Conversation with Gemini","depth":15,"bounds":{"left":0.09335937,"top":0.12847222,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Conversation with Gemini","depth":16,"bounds":{"left":0.09335937,"top":0.13055556,"width":0.14101562,"height":0.022222223},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":21,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\\nExpected: {gt_label}\\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.","depth":23,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Expand","depth":21,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Listen","depth":22,"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Gemini said","depth":20,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"This Python script defines the","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator","depth":25,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Core Functionality","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Core Functionality","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Data Collection (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"collect_test_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"* Supports fetching raw samples from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Elasticsearch (ES)","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or loading them from","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"local JSON files","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Implements a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"replayability check","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":": It probes the live API endpoint to ensure the activity exists before including it in the dataset.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Filters data based on request body size, language, and the presence of an original response.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Ground Truth Management:","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Saves inputs and \"gold standard\" responses to a local directory structure (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"eval_output/ai_activity_type/...","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":").","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Features an optional mode to populate ground truth using a high-order LLM (e.g.,","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"gpt-5_2","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") if the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"POPULATE_GROUND_TRUTH_LLM","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"flag is set.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Evaluation Logic (","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"evaluate_against_ground_truth","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Compares the model's predicted","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ai_activity_type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"against either the original ES response or the LLM-generated ground truth.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Assigns a","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"quality_score","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"of","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"1.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a match and","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"0.0","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for a mismatch or request failure.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Key Configuration Constants","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Key Configuration Constants","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Constant","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"ENDPOINT_PATH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"COMPARE_TO_GROUND_TRUTH","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"SKIP_UNREPLAYABLE_ON_COLLECT","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"REPLAYABILITY_CHECK_TIMEOUT_SECONDS","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Purpose","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The specific API route being tested:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"call/ai-activity-type","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Determines if the baseline is the original data (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"es","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") or an","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"llm","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"If","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"True","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", discards records where the API returns \"Cannot find activity.\"","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Limits the probe request to 20 seconds.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy table","depth":24,"bounds":{"left":0.22070312,"top":0.0,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"Execution Workflow","depth":23,"bounds":{"left":0.103125,"top":0.0,"width":0.13320312,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Execution Workflow","depth":24,"bounds":{"left":0.103125,"top":0.0,"width":0.060546875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Parsing:","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.024609376,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script uses","depth":26,"bounds":{"left":0.14179687,"top":0.009027778,"width":0.04609375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"create_parser","depth":27,"bounds":{"left":0.19023438,"top":0.010416667,"width":0.042578124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to handle CLI arguments like","depth":26,"bounds":{"left":0.1171875,"top":0.009027778,"width":0.119140625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--data-set","depth":27,"bounds":{"left":0.1171875,"top":0.028472222,"width":0.107421875,"height":0.03125},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.12929687,"top":0.045138888,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--populate-ground-truth-llm","depth":27,"bounds":{"left":0.13476562,"top":0.046527777,"width":0.08867188,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", and","depth":26,"bounds":{"left":0.1171875,"top":0.045138888,"width":0.11171875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"--compare-to-ground-truth","depth":27,"bounds":{"left":0.13203125,"top":0.06458333,"width":0.08203125,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":26,"bounds":{"left":0.21640626,"top":0.063194446,"width":0.0015625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Initialization:","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.039453126,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sets up directory paths for inputs and ground truths based on the selected dataset name.","depth":26,"bounds":{"left":0.1171875,"top":0.08680555,"width":0.11875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Run:","depth":26,"bounds":{"left":0.1171875,"top":0.14652778,"width":0.013671875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Executes the evaluation via","depth":26,"bounds":{"left":0.13085938,"top":0.14652778,"width":0.080078125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.run(run_evaluation(...))","depth":27,"bounds":{"left":0.11953125,"top":0.16597222,"width":0.10507812,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", which orchestrates the collection, endpoint calling, and result scoring phases.","depth":26,"bounds":{"left":0.1171875,"top":0.16458334,"width":0.11289062,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Copy prompt","depth":21,"bounds":{"left":0.1203125,"top":0.28194445,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Edit","depth":21,"bounds":{"left":0.1375,"top":0.28194445,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXHeading","text":"You said are there soem issues with this PR","depth":21,"bounds":{"left":0.1609375,"top":0.29027778,"width":0.072265625,"height":0.03888889},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"You said","depth":23,"bounds":{"left":0.09335937,"top":0.2923611,"width":0.0234375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"are there soem issues with this PR","depth":23,"bounds":{"left":0.1609375,"top":0.29305556,"width":0.062109374,"height":0.034027778},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXButton","text":"Listen","depth":22,"bounds":{"left":0.22382812,"top":0.35694444,"width":0.015625,"height":0.027777778},"help_text":"","role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXButton","text":"Show more options","depth":20,"bounds":{"left":0.22382812,"top":0.35208333,"width":0.015625,"height":0.027777778},"role_description":"button","subrole":"AXUnknown","is_enabled":true,"is_focused":false,"is_selected":false},{"role":"AXStaticText","text":"Show thinking","depth":26,"bounds":{"left":0.11953125,"top":0.35902777,"width":0.036328126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Gemini said","depth":20,"bounds":{"left":0.11679687,"top":0.3951389,"width":0.000390625,"height":0.00069444446},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Gemini said","depth":21,"bounds":{"left":0.11679687,"top":0.39652777,"width":0.048046876,"height":0.016666668},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.","depth":24,"bounds":{"left":0.10390625,"top":0.40347221,"width":0.13242188,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Here is a breakdown of the most notable issues, categorized by impact:","depth":24,"bounds":{"left":0.10390625,"top":0.46875,"width":0.11640625,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Performance & Async Anti-Patterns","depth":23,"bounds":{"left":0.10390625,"top":0.5222222,"width":0.13242188,"height":0.016666668},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Performance & Async Anti-Patterns","depth":24,"bounds":{"left":0.10390625,"top":0.5236111,"width":0.10859375,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Sequential Requests Defeating Async (Lines 86-106):","depth":26,"bounds":{"left":0.11757813,"top":0.5465278,"width":0.11601563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inside","depth":26,"bounds":{"left":0.16289063,"top":0.56458336,"width":0.019921875,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"_collect_from_es","depth":27,"bounds":{"left":0.11992188,"top":0.58402777,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", the code loops through","depth":26,"bounds":{"left":0.11757813,"top":0.58263886,"width":0.10429688,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"raw_samples","depth":27,"bounds":{"left":0.14414063,"top":0.6020833,"width":0.0359375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"and","depth":26,"bounds":{"left":0.18242188,"top":0.6006944,"width":0.0140625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"await","depth":27,"bounds":{"left":0.19882813,"top":0.6020833,"width":0.01640625,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"s a","depth":26,"bounds":{"left":0.21757813,"top":0.6006944,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"session.post","depth":27,"bounds":{"left":0.11992188,"top":0.6201389,"width":0.039453126,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using","depth":26,"bounds":{"left":0.11757813,"top":0.61875,"width":0.11640625,"height":0.08680555},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp","depth":27,"bounds":{"left":0.18046875,"top":0.6923611,"width":0.023046875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"for asynchronous concurrency.","depth":26,"bounds":{"left":0.11757813,"top":0.6909722,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.009765625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Collect these coroutines into a list and use","depth":28,"bounds":{"left":0.1296875,"top":0.7326389,"width":0.09921875,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.gather()","depth":29,"bounds":{"left":0.16601562,"top":0.75208336,"width":0.05234375,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(ideally wrapped in an","depth":28,"bounds":{"left":0.1296875,"top":0.75069445,"width":0.09257813,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.Semaphore","depth":29,"bounds":{"left":0.13203125,"top":0.7881944,"width":0.055859376,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"to avoid overwhelming the endpoint) to run the replayability probes concurrently.","depth":28,"bounds":{"left":0.1296875,"top":0.78680557,"width":0.10546875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):","depth":26,"bounds":{"left":0.11757813,"top":0.84652776,"width":0.10976563,"height":0.03263889},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Standard synchronous file operations (","depth":26,"bounds":{"left":0.11757813,"top":0.8645833,"width":0.107421875,"height":0.050694443},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"with open(...)","depth":27,"bounds":{"left":0.12226562,"top":0.90208334,"width":0.045703124,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.1703125,"top":0.90069443,"width":0.003515625,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.dump","depth":27,"bounds":{"left":0.17617187,"top":0.90208334,"width":0.029296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":",","depth":26,"bounds":{"left":0.2078125,"top":0.90069443,"width":0.003125,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"json.load","depth":27,"bounds":{"left":0.11992188,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") are used inside","depth":26,"bounds":{"left":0.15195313,"top":0.91875,"width":0.048046876,"height":0.014583333},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"async def","depth":27,"bounds":{"left":0.20234375,"top":0.9201389,"width":0.0296875,"height":0.013194445},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"functions. In Python, this blocks the entire event loop while the disk is reading/writing.","depth":26,"bounds":{"left":0.11757813,"top":0.91875,"width":0.11875,"height":0.06875},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.009765625,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Use an asynchronous file I/O library like","depth":28,"bounds":{"left":0.1296875,"top":0.9965278,"width":0.09335937,"height":0.003472209},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiofiles","depth":29,"bounds":{"left":0.1625,"top":1.0,"width":0.0265625,"height":-0.015972257},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":", or run the blocking I/O operations in a separate thread using","depth":28,"bounds":{"left":0.1296875,"top":1.0,"width":0.10585938,"height":-0.014583349},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"asyncio.to_thread()","depth":29,"bounds":{"left":0.16914062,"top":1.0,"width":0.0625,"height":-0.052083373},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"bounds":{"left":0.23398438,"top":1.0,"width":0.0015625,"height":-0.050694466},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Architecture & State Management","depth":23,"bounds":{"left":0.10390625,"top":1.0,"width":0.13242188,"height":-0.08611107},"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Architecture & State Management","depth":24,"bounds":{"left":0.10390625,"top":1.0,"width":0.103515625,"height":-0.087499976},"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Mutating Class-Level Variables (Lines 323-328):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"The script parses CLI arguments and directly mutates the class attributes (e.g.,","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"AiActivityTypeEvaluator.DATA_SET = args.data_set","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Pass these arguments into the","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"__init__","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"constructor of the instance (","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"self.data_set = args.data_set","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":") rather than modifying the class definition globally.","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXHeading","text":"Code Quality & Edge Cases","depth":23,"help_text":"","role_description":"heading","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Code Quality & Edge Cases","depth":24,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Broad Exception Catching (Line 107):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Using a bare","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except Exception:","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"during the replayability probe is dangerous. It will catch","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"aiohttp.ClientError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"(which is intended), but it will also silently swallow","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"TimeoutError","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"or even memory errors, making debugging incredibly difficult if the endpoint goes down.","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Fix:","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Scope the exception to specific expected errors, like","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"except (aiohttp.ClientError, asyncio.TimeoutError):","depth":29,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":".","depth":28,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"Inefficient Size Calculation (Line 95):","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"len(json.dumps(request_body)) < min_size","depth":27,"help_text":"","role_description":"text","subrole":"AXUnknown"},{"role":"AXStaticText","text":"requires serializing the entire dictionary to a string","depth":26,"help_text":"","role_description":"text","subrole":"AXUnknown"}]...
|
584939840227356054
|
6138248189675614487
|
visual_change
|
accessibility
|
NULL
|
Platform Sprint 1 Q2 - Platform Team - Scrum Board Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
jiminny.atlassian.net
JY-20543 add AJ reports User pilot tracking by LakyLak · Pull Request #11932 · jiminny/app
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Platform Sprint 1 Q2 - Platform Team - Scrum Board - Jira
Close tab
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jy 19798 evaluation for ai activity types by nikolaybiaivanov · Pull Request #468 · jiminny/prophet
Jiminny
Jiminny
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Ask Jiminny test report - 8 Apr 2026 - Ask Jiminny test report - 13 Apr 2026.pdf
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Your Ask Jiminny test report report is ready - [EMAIL] - Jiminny Mail
Close tab
New Tab
Customize sidebar
Close Google Gemini (⌃X)
Tabs from other devices
Open history (⇧⌘H)
Open bookmarks (⌘B)
AI Chat settings
Close
WORK, Google Account: [EMAIL]
Main menu
New Chat
Open menu for conversation actions.
Conversation with Gemini
Conversation with Gemini
Copy prompt
You said I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected. Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
You said
I’m on page “<tabTitle>Jy 19798 evaluation for ai activity types by nikol</tabTitle>” with “<selection>@@ -0,0 +1,330 @@1+#!/usr/bin/env python32+import asyncio3+import json4+import os5+from pathlib import Path6+from typing import Any, Dict, List7+8+import aiohttp9+10+from src.component.call_processing.schemes.ai_activity_type_request import AiActivityTypeRequest11+from src.component.call_processing.schemes.ai_activity_type_response import AiActivityTypeResponse12+from src.eval.abstract_evaluator import AbstractEvaluator13+from src.eval.collectors import collect_test_set as es_collect_test_set14+from src.eval.common_evaluator_runner import create_parser, run_evaluation15+from src.eval.endpoints import run_endpoint as run_endpoint_http16+from src.eval.models import EvaluationResult17+18+19+class AiActivityTypeEvaluator(AbstractEvaluator):20+ENDPOINT_PATH = "call/ai-activity-type"21+REQUEST_MODEL = AiActivityTypeRequest22+RESPONSE_MODEL = AiActivityTypeResponse23+OUTPUT_SUBFOLDER = "ai_activity_type"24+USES_LLM_EVALUATION = False25+26+READS_FROM_DATASET_FOLDER = True27+28+DATA_SET: str = "initial"29+30+MINIMUM_REQUEST_BODY_SIZE: int = 031+32+SKIP_UNREPLAYABLE_ON_COLLECT: bool = True33+REPLAYABILITY_CHECK_TIMEOUT_SECONDS: float = 20.034+35+POPULATE_GROUND_TRUTH_LLM: bool = False36+GROUND_TRUTH_LLM_ENDPOINT: str = "OPENAI:gpt-5_2"37+38+COMPARE_TO_GROUND_TRUTH: str = "es"39+40+def __init__(self, **kwargs: Any) -> None:41+super().__init__(**kwargs)42+_repo_root = Path(__file__).parents[2]43+_datasets_root = _repo_root / "eval_output" / self.OUTPUT_SUBFOLDER / "eval_datasets"44+_dataset_dir = _datasets_root / self.__class__.DATA_SET45+self._dataset_dir = _dataset_dir46+self._inputs_dir = _dataset_dir / "input"47+self._ground_truths_dir = _dataset_dir / "ground_truth"48+49+async def collect_test_set(self) -> List[Dict[str, Any]]:50+if self.es_host_name:51+return await self._collect_from_es()52+return await self._collect_from_files()53+54+async def _collect_from_es(self) -> List[Dict[str, Any]]:55+if self._inputs_dir.exists() or self._ground_truths_dir.exists():56+raise FileExistsError(57+f"Dataset folder already exists: {self._inputs_dir.parent}. "58+"Choose a different --data-set name or remove the folder manually."59+ )60+61+raw_samples = await es_collect_test_set(62+es_host_name=self.es_host_name,63+es_port=self.es_port,64+endpoint_path=self.ENDPOINT_PATH,65+limit=self.limit,66+language_filter=self.language_filter,67+min_prompt_length=self.min_prompt_length,68+is_json_response=self.IS_JSON_RESPONSE,69+exclude_languages=self.exclude_languages,70+ )71+72+self._inputs_dir.mkdir(parents=True, exist_ok=True)73+self._ground_truths_dir.mkdir(parents=True, exist_ok=True)74+75+min_size = self.MINIMUM_REQUEST_BODY_SIZE76+samples: List[Dict[str, Any]] = []77+skipped_no_response = 078+skipped_too_small = 079+skipped_unreplayable = 080+skipped_probe_errors = 081+82+endpoint_url = f"{self.api_host}/{self.ENDPOINT_PATH}"83+84+timeout = aiohttp.ClientTimeout(total=self.REPLAYABILITY_CHECK_TIMEOUT_SECONDS)85+async with aiohttp.ClientSession(timeout=timeout) as session:86+for sample in raw_samples:87+es_id = sample["id"]88+request_body = sample["request_body"]89+original_response = sample.get("original_response")90+91+if not original_response:92+skipped_no_response += 193+continue94+95+if min_size and len(json.dumps(request_body)) < min_size:96+skipped_too_small += 197+continue98+99+if self.SKIP_UNREPLAYABLE_ON_COLLECT:100+try:101+async with session.post(endpoint_url, json=request_body) as resp:102+if resp.status != 200:103+body_text = await resp.text()104+if "Cannot find activity" in body_text:105+skipped_unreplayable += 1106+continue107+except Exception:108+skipped_probe_errors += 1109+110+with open(self._inputs_dir / f"{es_id}.json", "w") as f:111+json.dump(request_body, f, indent=2)112+113+with open(self._ground_truths_dir / f"gt_{es_id}.json", "w") as f:114+json.dump(original_response, f, indent=2)115+116+samples.append(117+ {118+"id": es_id,119+"request_body": request_body,120+"original_response": original_response,121+"ground_truth": original_response,122+ }123+ )124+125+print(f"Wrote {len(samples)} samples to {self._inputs_dir.parent}")126+if skipped_no_response:127+print(f"Skipped {skipped_no_response} records with no response body")128+if skipped_too_small:129+print(130+f"Skipped {skipped_too_small} records whose request body was smaller "131+f"than {min_size} characters"132+ )133+if skipped_unreplayable:134+print(f"Skipped {skipped_unreplayable} records that are not replayable (missing activity in ES)")135+if skipped_probe_errors:136+print(f"Warning: replayability probe errored for {skipped_probe_errors} records (kept those records)")137+138+if self.__class__.POPULATE_GROUND_TRUTH_LLM:139+label_results = await run_endpoint_http(140+test_set=samples,141+endpoint_url=endpoint_url,142+request_model=self.REQUEST_MODEL,143+force_llm_endpoint=self.__class__.GROUND_TRUTH_LLM_ENDPOINT,144+is_streaming_response=self.IS_STREAMING_RESPONSE,145+timeout_seconds=300.0,146+parallel_requests=self.parallel_requests,147+return_prompt_in_run=self.RETURN_PROMPT_IN_RUN,148+ )149+150+for i, r in enumerate(label_results):151+forced = self.__class__.GROUND_TRUTH_LLM_ENDPOINT152+model = forced.split(":", 1)[1] if ":" in forced else forced153+samples[i]["ground_truth_llm_model"] = model154+155+if not r.get("success", True):156+samples[i]["ground_truth_llm"] = None157+samples[i]["ground_truth_llm_reasoning"] = None158+samples[i]["ground_truth_llm_error_type"] = r.get("error_type")159+samples[i]["ground_truth_llm_error_message"] = r.get("error_message")160+continue161+162+resp = r.get("new_response") or {}163+samples[i]["ground_truth_llm"] = resp.get("ai_activity_type")164+samples[i]["ground_truth_llm_reasoning"] = resp.get("ai_activity_type_reasoning")165+166+return samples167+168+async def _collect_from_files(self) -> List[Dict[str, Any]]:169+if not self._inputs_dir.exists():170+raise FileNotFoundError(171+f"Dataset folder not found: {self._inputs_dir}. "172+"Run --steps collect first or check your --data-set name."173+ )174+175+samples: List[Dict[str, Any]] = []176+for filename in sorted(os.listdir(self._inputs_dir)):177+input_path = self._inputs_dir / filename178+gt_path = self._ground_truths_dir / f"gt_{filename}"179+180+if not gt_path.exists():181+print(f"Warning: ground truth not found for {filename}, skipping")182+continue183+184+with open(input_path, "r") as f:185+request_body: Dict[str, Any] = json.load(f)186+187+with open(gt_path, "r") as f:188+ground_truth: Dict[str, Any] = json.load(f)189+190+samples.append(191+ {192+"id": Path(filename).stem,193+"request_body": request_body,194+"original_response": None,195+"ground_truth": ground_truth,196+ }197+ )198+199+print(f"Loaded {len(samples)} samples from {self._inputs_dir.parent}")200+return samples201+202+async def evaluate_against_ground_truth(self, results: List[Dict[str, Any]]) -> List[EvaluationResult]:203+evaluation_results: List[EvaluationResult] = []204+compare_to = self.__class__.COMPARE_TO_GROUND_TRUTH205+206+for res in results:207+req_id = res["request_id"]208+is_success = res.get("success", True)209+210+if not is_success:211+evaluation_results.append(212+EvaluationResult(213+request_id=req_id,214+request_body=res["request_body"],215+new_response=res.get("new_response"),216+valid_schema=False,217+quality_score=0.0,218+quality_feedback=f"Request failed: {res.get('error_type')} - {res.get('error_message')}",219+response_time=res.get("response_time"),220+success=False,221+error_type=res.get("error_type"),222+error_message=res.get("error_message"),223+ground_truth_match=False,224+ )225+ )226+continue227+228+pred = (res.get("new_response") or {}).get("ai_activity_type")229+230+if compare_to == "llm":231+if "ground_truth_llm" not in res:232+evaluation_results.append(233+EvaluationResult(234+request_id=req_id,235+request_body=res["request_body"],236+new_response=res.get("new_response"),237+valid_schema=res.get("valid_schema", True),238+quality_score=0.0,239+quality_feedback=(240+"Missing ground_truth_llm in run results. Ensure collect step populated it and "241+"endpoints.py passes it through."242+ ),243+response_time=res.get("response_time"),244+success=True,245+ground_truth_match=False,246+ )247+ )248+continue249+gt_label = res.get("ground_truth_llm")250+else:251+gt_path = self._ground_truths_dir / f"gt_{req_id}.json"252+if not gt_path.exists():253+evaluation_results.append(254+EvaluationResult(255+request_id=req_id,256+request_body=res["request_body"],257+new_response=res.get("new_response"),258+valid_schema=res.get("valid_schema", True),259+quality_score=0.0,260+quality_feedback=f"Ground truth file not found: {gt_path}",261+response_time=res.get("response_time"),262+success=True,263+ground_truth_match=False,264+ )265+ )266+continue267+268+with open(gt_path, "r") as f:269+ground_truth_dict: Dict[str, Any] = json.load(f)270+271+gt_label = (ground_truth_dict or {}).get("ai_activity_type")272+273+matched = pred == gt_label274+275+evaluation_results.append(276+EvaluationResult(277+request_id=req_id,278+request_body=res["request_body"],279+new_response=res.get("new_response"),280+valid_schema=res.get("valid_schema", True),281+quality_score=1.0 if matched else 0.0,282+quality_feedback="PASS" if matched else f"FAIL\nExpected: {gt_label}\nGot: {pred}",283+response_time=res.get("response_time"),284+success=True,285+ground_truth_match=matched,286+ )287+ )288+289+return evaluation_results290+291+292+if __name__ == "__main__":293+parser = create_parser(AiActivityTypeEvaluator)294+parser.add_argument(295+"--data-set",296+type=str,297+default="initial",298+dest="data_set",299+help="Dataset folder under eval_output/ai_activity_type/eval_datasets/ (default: initial)",300+ )301+302+parser.add_argument(303+"--populate-ground-truth-llm",304+action="store_true",305+dest="populate_ground_truth_llm",306+ )307+parser.add_argument(308+"--ground-truth-llm-endpoint",309+type=str,310+default="OPENAI:gpt-5_2",311+dest="ground_truth_llm_endpoint",312+ )313+314+parser.add_argument(315+"--compare-to-ground-truth",316+type=str,317+choices=["es", "llm"],318+default="es",319+dest="compare_to_ground_truth",320+ )321+322+args = parser.parse_args()323+AiActivityTypeEvaluator.DATA_SET = args.data_set324+325+AiActivityTypeEvaluator.POPULATE_GROUND_TRUTH_LLM = bool(args.populate_ground_truth_llm)326+AiActivityTypeEvaluator.GROUND_TRUTH_LLM_ENDPOINT = args.ground_truth_llm_endpoint327+328+AiActivityTypeEvaluator.COMPARE_TO_GROUND_TRUTH = args.compare_to_ground_truth329+330+asyncio.run(run_evaluation(AiActivityTypeEvaluator, args))</selection>” selected.
Please summarize the selection using precise and concise language. Use headers and bulleted lists in the summary, to make it scannable. Maintain the meaning and factual accuracy.
Expand
Listen
Show more options
Gemini said
Gemini said
This Python script defines the
AiActivityTypeEvaluator
class, a specialized tool for evaluating AI-driven activity type classification. It extends a base evaluation framework to handle data collection, ground-truth comparison, and performance scoring.
Core Functionality
Core Functionality
Data Collection (
collect_test_set
):
* Supports fetching raw samples from
Elasticsearch (ES)
or loading them from
local JSON files
.
Implements a
replayability check
: It probes the live API endpoint to ensure the activity exists before including it in the dataset.
Filters data based on request body size, language, and the presence of an original response.
Ground Truth Management:
Saves inputs and "gold standard" responses to a local directory structure (
eval_output/ai_activity_type/...
).
Features an optional mode to populate ground truth using a high-order LLM (e.g.,
gpt-5_2
) if the
POPULATE_GROUND_TRUTH_LLM
flag is set.
Evaluation Logic (
evaluate_against_ground_truth
):
Compares the model's predicted
ai_activity_type
against either the original ES response or the LLM-generated ground truth.
Assigns a
quality_score
of
1.0
for a match and
0.0
for a mismatch or request failure.
Key Configuration Constants
Key Configuration Constants
Constant
Purpose
ENDPOINT_PATH
The specific API route being tested:
call/ai-activity-type
.
COMPARE_TO_GROUND_TRUTH
Determines if the baseline is the original data (
es
) or an
llm
.
SKIP_UNREPLAYABLE_ON_COLLECT
If
True
, discards records where the API returns "Cannot find activity."
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Limits the probe request to 20 seconds.
Constant
ENDPOINT_PATH
COMPARE_TO_GROUND_TRUTH
SKIP_UNREPLAYABLE_ON_COLLECT
REPLAYABILITY_CHECK_TIMEOUT_SECONDS
Purpose
The specific API route being tested:
call/ai-activity-type
.
Determines if the baseline is the original data (
es
) or an
llm
.
If
True
, discards records where the API returns "Cannot find activity."
Limits the probe request to 20 seconds.
Copy table
Execution Workflow
Execution Workflow
Parsing:
The script uses
create_parser
to handle CLI arguments like
--data-set
,
--populate-ground-truth-llm
, and
--compare-to-ground-truth
.
Initialization:
Sets up directory paths for inputs and ground truths based on the selected dataset name.
Run:
Executes the evaluation via
asyncio.run(run_evaluation(...))
, which orchestrates the collection, endpoint calling, and result scoring phases.
Copy prompt
Edit
You said are there soem issues with this PR
You said
are there soem issues with this PR
Listen
Show more options
Show thinking
Gemini said
Gemini said
Yes, there are several issues and code smells in this Pull Request, ranging from performance bottlenecks to architectural anti-patterns.
Here is a breakdown of the most notable issues, categorized by impact:
Performance & Async Anti-Patterns
Performance & Async Anti-Patterns
Sequential Requests Defeating Async (Lines 86-106):
Inside
_collect_from_es
, the code loops through
raw_samples
and
await
s a
session.post
request for each sample one by one. This means the script waits for one HTTP request to finish before starting the next, completely defeating the purpose of using
aiohttp
for asynchronous concurrency.
Fix:
Collect these coroutines into a list and use
asyncio.gather()
(ideally wrapped in an
asyncio.Semaphore
to avoid overwhelming the endpoint) to run the replayability probes concurrently.
Blocking File I/O in Async Functions (Lines 110-114, 184-188, 268-269):
Standard synchronous file operations (
with open(...)
,
json.dump
,
json.load
) are used inside
async def
functions. In Python, this blocks the entire event loop while the disk is reading/writing.
Fix:
Use an asynchronous file I/O library like
aiofiles
, or run the blocking I/O operations in a separate thread using
asyncio.to_thread()
.
Architecture & State Management
Architecture & State Management
Mutating Class-Level Variables (Lines 323-328):
The script parses CLI arguments and directly mutates the class attributes (e.g.,
AiActivityTypeEvaluator.DATA_SET = args.data_set
). This is a major anti-pattern. If you ever try to run tests in parallel or create multiple instances of this evaluator in a single process, the state will bleed across them.
Fix:
Pass these arguments into the
__init__
constructor of the instance (
self.data_set = args.data_set
) rather than modifying the class definition globally.
Code Quality & Edge Cases
Code Quality & Edge Cases
Broad Exception Catching (Line 107):
Using a bare
except Exception:
during the replayability probe is dangerous. It will catch
aiohttp.ClientError
(which is intended), but it will also silently swallow
TimeoutError
or even memory errors, making debugging incredibly difficult if the endpoint goes down.
Fix:
Scope the exception to specific expected errors, like
except (aiohttp.ClientError, asyncio.TimeoutError):
.
Inefficient Size Calculation (Line 95):
len(json.dumps(request_body)) < min_size
requires serializing the entire dictionary to a string...
|
NULL
|