Fix/multiclass prediction type safety

[gbeane]

This pull request introduces improvements to how classifier settings are scoped by behavior and adds robust validation to prediction loading to prevent mismatches between binary and multi-class prediction records. It also updates the relevant tests to ensure correct behavior in these scenarios.

Classifier settings scoping improvements:

• The Classifier.set_project_settings method now accepts an optional behavior argument, allowing explicit scoping of settings without relying on the classifier's behavior_name attribute. This makes the method more flexible and avoids previous ordering requirements. (src/jabs/classifier/classifier.py [1] src/jabs/classifier/cross_validation.py [2]
• The MultiClassClassifier.set_project_settings method signature is updated for consistency, accepting a (unused) behavior argument to match the base class, simplifying interface usage. (src/jabs/classifier/multi_class_classifier.py src/jabs/classifier/multi_class_classifier.pyL121-R130)
• Corresponding test cases are added to verify that explicitly provided behavior arguments are preferred over the attribute and that the correct project settings are loaded. (tests/classifier/test_classifier.py [1] tests/classifier/test_cross_validation.py [2]

Prediction loading validation:

• The prediction manager now validates that the mode (binary vs multi-class) of loaded prediction records matches the caller's expectations. If a mismatch is detected, a ValueError is raised instead of silently returning mis-shaped data, preventing subtle bugs. (src/jabs/project/prediction_manager.py [1] [2] [3]
• New tests ensure that attempts to load a multi-class record as binary (or vice versa) are correctly rejected with errors, rather than returning incorrect data. (tests/project/test_prediction_manager.py tests/project/test_prediction_manager.pyR231-R284)

[Copilot]

Pull request overview

Improves type safety around prediction loading by validating the binary/multi-class mode of stored records and refactors Classifier.set_project_settings to accept an explicit behavior argument, removing the prior requirement to set behavior_name first.

Changes:

• Classifier.set_project_settings gains optional behavior arg; MultiClassClassifier mirrors the signature for parity; cross-validation now passes behavior explicitly.
• PredictionManager._load_prediction_record requires keyword expect_multiclass and raises ValueError when the stored record's mode (presence of class_names) doesn't match.
• New tests cover explicit-behavior precedence and cross-mode prediction rejection.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/jabs/classifier/classifier.py	Accept optional behavior arg; prefer it over self._behavior when scoping settings.
src/jabs/classifier/multi_class_classifier.py	Add unused behavior parameter for signature parity, with docstring.
src/jabs/classifier/cross_validation.py	Pass behavior explicitly to set_project_settings and reorder relative to behavior_name assignment.
src/jabs/project/prediction_manager.py	Add expect_multiclass keyword and raise ValueError on mode mismatch; expanded docstring.
tests/classifier/test_classifier.py	New tests covering explicit behavior overriding/preferring over attribute.
tests/classifier/test_cross_validation.py	Update fake classifier signature to match new set_project_settings.
tests/project/test_prediction_manager.py	New tests verifying cross-mode reads raise ValueError.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.