fix(augur): use regression-appropriate scorers for regressor estimators

[Sizerta]

PR Checklist

PR Checklist

• Referenced issue is linked
• If you've fixed a bug or added code that should be tested, add tests!

Description of changes

Fixes #655

Augur with a regressor estimator (e.g. random_forest_regressor) could crash during predict() with

AttributeError: RandomForestRegressor has none of the following attributes: predict_proba.

when the target had more than two distinct values (for example, multiple timepoints or a continuous outcome). In this case, the multiclass branch of set_scorer was selected and returned multiclass classification scorers, including scorers using response_method="predict_proba", even though the estimator was a regressor.

The non-multiclass fallback also used a hardcoded RandomForestClassifier | LogisticRegression whitelist to select classification scorers. As a result, classifiers outside that whitelist could receive regression scorers.

set_scorer now checks whether the estimator is a regressor first, using scikit-learn's is_regressor():

1. Regressors → regression scorers (augur_score, r2, ccc, neg_mean_squared_error, explained_variance), regardless of the multiclass argument.
2. Classifiers with multiclass=True → multiclass classification scorers (unchanged).
3. Other non-regressor estimators → binary classification scorers (unchanged scoring behavior, without the previous estimator whitelist).

This prevents regressor-based Augur runs from requesting class probabilities and ensures classifiers outside the previous hardcoded whitelist follow the classification-scoring path.

Technical details

• src/pertpy/tools/_augur.py: reordered set_scorer so the is_regressor(self.estimator) branch is evaluated before the multiclass branch, and replaced the previous estimator-type ternary with the standard classification-scoring path.

• tests/tools/test_augur.py: added two regression tests for Got an error when using 'random_forest_regressor' estimator #655:

  • test_regressor_scorer_with_more_than_two_labels — verifies that set_scorer(multiclass=True) returns the regression scorer set for a random-forest regressor.
  • test_regressor_cross_validation_with_four_timepoints — end-to-end test using four numeric timepoints, verifying that cross-validation produces a finite mean_ccc and does not produce mean_auc.

The regression-specific scorer test ensures that a regressor with multiple target values is not assigned classification scorers. The end-to-end test verifies that the corrected scorer selection avoids the predict_proba() failure during cross-validation.

Environment: Windows, Python 3.14.6, scikit-learn 1.9.1. The mypy pre-commit hook and the full hatch test matrix could not run locally because ete4 (via the tcoda extra) fails to build with Cython on Windows; this is pre-existing and unrelated to this change. CI will cover the supported test matrix.

Additional context

multiclass is only meaningful for classifiers. A regressor with more than two distinct numeric target values remains a regression problem, so the estimator type must take precedence over the multiclass flag.

The documentation checklist item from the original template was removed because this is a bug fix with no API change requiring documentation updates.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.97%. Comparing base (e8abf85) to head (7b108ab).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1105   +/-   ##
=======================================
  Coverage   79.97%   79.97%           
=======================================
  Files          55       55           
  Lines        7534     7536    +2     
=======================================
+ Hits         6025     6027    +2     
  Misses       1509     1509           

Files with missing lines	Coverage Δ
src/pertpy/tools/_augur.py	89.66% <100.00%> (+0.06%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Zethson]

Great, thank you!