set classifier metadata

[sarptandoven]

summary

sets the sklearn classifier metadata that NGBClassifier was missing, so sklearn tooling can treat it as a classifier after fitting.

fixes #350.

what changed

• make NGBClassifier inherit from ClassifierMixin
• set classes_ during fit and first partial_fit
• encode labels internally with LabelEncoder
• decode predict and staged_predict back to the original labels
• encode Y_val when validation data is passed
• keep predict_proba aligned with classes_
• add classifier regression coverage for sklearn compatibility

why

sklearn metrics/display helpers such as RocCurveDisplay.from_estimator and cross_val_score(..., scoring="roc_auc") expect fitted classifiers to expose classifier metadata, especially classes_.

without it, NGBClassifier can fit and predict, but fails in common sklearn evaluation paths.

validation

• git diff --check
• make lint
• manual sklearn probe for issue 'NGBClassifier' object has no attribute 'classes_' #350, string labels, validation labels, partial_fit, multiclass labels, clone, and pickle round-trip
• pytest tests/test_basic.py::test_classifier_sets_sklearn_classes_and_encodes_labels -q
• pytest tests/test_basic.py -q
• pytest -q
• make test

[ryan-wolbeck]

Thanks for the update. #407 has now merged into master, but this PR is still based on the pre-#407 commit, so GitHub’s diff is showing some stale overlap from #407 — specifically the NGBClassifier validation_fraction / early_stopping_rounds API changes.

Could you please rebase this branch onto latest master and force-push with lease?