Overhaul Python PCA, from NaN handling to SKLearn replacement#2311

Open

jucor wants to merge 29 commits intocompdemocracy:edgefrom

jucor:pca_work

Commits on Jan 6, 2026

Squashed commit of the following:
jucor
committed
Make each test run their own Conversation object
jucor
committed
Actually reuse the fixtures...
jucor
committed
Enable parallel test execution by dataset using xdist_group markers

jucor
and
claude
committed
Factorize xdist_group pattern across test suite for parallel execution

jucor
and
claude
committed
Add --datasets option to run tests on specific datasets

jucor
and
claude
committed
Fix --datasets filter for test_legacy_clojure_regression.py

jucor
and
claude
committed
Add normalized angle computation to PCA regression test

jucor
and
claude
committed
Remove madness "Align with clojure" and update golden records
jucor
committed
Save difference log regardless of errors
jucor
committed
Account for flip sign in numerical difference reporting
jucor
committed
Add PCA benchmark script
jucor
committed
Replace handwritten PCA by SKLearn's implementation
jucor
committed
Replace projection by Sklearn's fit_transform for better performance.
jucor
committed
Remove unused start_vectors argument
jucor
committed
Remove function wrapped_pca
jucor
committed
Reformat help message
jucor
committed
Add cluster centers to PCA-related paths for sign flip handling

jucor
and
claude
committed
Allow for a few components with higher error tolerance
jucor
committed
Fix per-component PCA sign flip detection

jucor
and
claude
committed
Replace element-wise tolerance with projection metrics for regression tests
jucor
committed
Update biodiversity and vw golden snapshots with sklearn PCA
jucor
committed
Fix PCA warnings in test_conversation.py

jucor
and
claude
committed
Replace PCA Clojure comparison test with pass

jucor
and
claude
committed
Consolidate Clojure comparison tests into test_legacy_clojure_regression.py

jucor
and
claude
committed
Address Copilot review feedback on PCA implementation

jucor
and
claude
committed
Document parallel test execution with pytest-xdist

jucor
and
claude
committed

Commits on Jan 7, 2026

FIx path of mock file for test run_math_pipeline_e2e
jucor
committed
Clarify test name
jucor
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Overhaul Python PCA, from NaN handling to SKLearn replacement#2311

Overhaul Python PCA, from NaN handling to SKLearn replacement#2311
jucor wants to merge 29 commits intocompdemocracy:edgefrom
jucor:pca_work

Commits on Jan 6, 2026

Squashed commit of the following:

Make each test run their own Conversation object

Actually reuse the fixtures...

Enable parallel test execution by dataset using xdist_group markers

Factorize xdist_group pattern across test suite for parallel execution

Add --datasets option to run tests on specific datasets

Fix --datasets filter for test_legacy_clojure_regression.py

Add normalized angle computation to PCA regression test

Remove madness "Align with clojure" and update golden records

Save difference log regardless of errors

Account for flip sign in numerical difference reporting

Add PCA benchmark script

Replace handwritten PCA by SKLearn's implementation

Replace projection by Sklearn's fit_transform for better performance.

Remove unused start_vectors argument

Remove function wrapped_pca

Reformat help message

Add cluster centers to PCA-related paths for sign flip handling

Allow for a few components with higher error tolerance

Fix per-component PCA sign flip detection

Replace element-wise tolerance with projection metrics for regression tests

Update biodiversity and vw golden snapshots with sklearn PCA

Fix PCA warnings in test_conversation.py

Replace PCA Clojure comparison test with pass

Consolidate Clojure comparison tests into test_legacy_clojure_regression.py

Address Copilot review feedback on PCA implementation

Document parallel test execution with pytest-xdist

Commits on Jan 7, 2026

FIx path of mock file for test run_math_pipeline_e2e

Clarify test name