[ChrisW09]

Fixes #77.

RepresentationSearchCV around a classifier built every candidate Preprocessor with the default task="regression". Target-aware candidates ("ple", and the feature maps and splines that Preprocessor places on the target by default) crashed on string class labels, and they placed bins, centers and knots for integer labels with a regression model.

Root cause

RepresentationSearchCV already checks is_classifier(self.estimator) to choose a stratified splitter, but _make_preprocessor forwarded only random_state and preprocessor_params, so it never derived task from the estimator.

Fix

RepresentationSearchCV._make_preprocessor sets task="classification" with params.setdefault when the estimator is a classifier. A task given in preprocessor_params still takes precedence.

Behaviour changes

• A search around a classifier fits its preprocessors with task="classification": string labels work, integer labels are placed by a classification model (DecisionTreeClassifier for "cart", LightGBM's classification objective for "lightgbm"), and best_preprocessor_.task is "classification". For binary 0/1 labels with "cart" placement the locations stay the same, as the issue explains. Multiclass labels and "lightgbm" placement can get different locations.
• Searches around a regressor, and searches that set task in preprocessor_params, are unchanged.

Documentation

• RepresentationSearchCV docstring (preprocessor_params): with a classifier as estimator, task defaults to "classification". The private _make_preprocessor docstring says the same.
• docs/core_concepts/target_awareness.md ("Searching over representations"): with a classifier, every Preprocessor uses task="classification", so target-aware methods place against the class labels, and a task set in preprocessor_params takes precedence.

Tests

• test_classifier_search_places_against_class_labels: a search around LogisticRegression on string labels with methods ["minmax", "ple"] fits, best_preprocessor_.task == "classification", and the predictions are class labels. Fails on main (could not convert string to float: np.str_('pos')), passes here.
• test_classifier_search_keeps_an_explicit_task: preprocessor_params={"task": "regression"} overrides the classifier default. Guard test: passes on both by design.
• test_regressor_search_keeps_the_regression_task: a search around a regressor keeps task="regression". Guard test: passes on both by design.

Reproduction

Output of the script from #77 on this branch:

string labels -> best_preprocessor_.task = classification
integer labels -> best_preprocessor_.task = classification

Checks

• just test: 1441 passed, 59 skipped, 7 xfailed; coverage 93.65% (gate 90%).
• Minimum dependencies (CI job: Python 3.10, numpy 1.24.4, pandas 2.0.3, scipy 1.10.1, scikit-learn 1.6.0): 1428 passed, 72 skipped, 7 xfailed; scripts/quickstart.py passes.
• just docs: strict build succeeds.
• just check: all pre-commit and pre-push hooks pass (ruff format, ruff lint, prettier, file hygiene, pyright with 0 errors).
• Import direction between subpackages is unchanged.

Merge notes

• Base: main.
• The PRs for RepresentationSearchCV cannot use group-based CV splitters and accepts no groups or fit parameters #91 (fix(search): pass groups to the splitter and fit params to the estimator #126) and RepresentationSearchCV lacks predict_proba, classes_ and classifier tags, so nested roc_auc CV returns NaN #92 (fix(search): make the search a classifier when it wraps a classifier #127) also change pretab/compose/search.py and docs/core_concepts/target_awareness.md. Both files merge cleanly with this branch. tests/compose/test_search.py has an append/append conflict at the end of the file: keep both blocks. RepresentationSearchCV lacks predict_proba, classes_ and classifier tags, so nested roc_auc CV returns NaN #92 adds the same class_data fixture, which git merges into one definition. Combined with either branch this way, the tests/compose suite passes. After this PR, the explicit {"task": "classification"} in RepresentationSearchCV lacks predict_proba, classes_ and classifier tags, so nested roc_auc CV returns NaN #92's tests is redundant but harmless.
• The PR for An invalid task such as "Regression" is silently treated as classification by the splines and selectors #93 (fix(validation): reject an unknown task instead of using classification #128) (task validation) merges cleanly. "classification" is a valid task there.
• With preprocessor_params={"placement_strategy": "lightgbm"}, a classifier search on string labels still fails on this branch alone, because LightGBM placement uses the binary objective on the raw labels. It works once the PR for LightGBM placement with task="classification" always uses the binary objective on raw labels #61 (fix(placement): label-encode classification targets for LightGBM #105) is merged too (checked by combining the two branches).

This change was previously part of #86, which is closed in favour of one pull request per issue.