fix: MultiOutputClassifier exposes classes_ (swev-id: scikit-learn__scikit-learn-14629)

[casey-brooks]

Summary

• add wrapper-level classes_ to MultiOutputClassifier and refresh it during partial_fit
• document classes_ and add regression coverage for cross_val_predict with MultiOutputClassifier
• ensure ClassifierChain order validation keeps expected indentation

Testing

• . /workspace/sklearn-py37/bin/activate && PYTHONPATH=. pytest sklearn/model_selection/tests/test_validation.py -k multioutput -vv
• . /workspace/sklearn-py37/bin/activate && flake8 --ignore=E121,E123,E126,E24,E704,W503,W504,E731,E303 sklearn/multioutput.py sklearn/model_selection/tests/test_validation.py
• . /workspace/sklearn-py37/bin/activate && PYTHONPATH=. python - <<'PY'
  from sklearn.datasets import make_multilabel_classification
  from sklearn.multioutput import MultiOutputClassifier
  from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
  from sklearn.model_selection import cross_val_predict
  X, Y = make_multilabel_classification(random_state=0)
  mo_lda = MultiOutputClassifier(LinearDiscriminantAnalysis())
  cross_val_predict(mo_lda, X, Y, cv=5, method='predict_proba')
  PY

Failure stack trace (before fix)

Traceback (most recent call last):
  File "<stdin>", line 9, in <module>
  File "/workspace/scikit-learn/sklearn/model_selection/_validation.py", line 766, in cross_val_predict
    for train, test in cv.split(X, y, groups))
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/parallel.py", line 1029, in __call__
    if self.dispatch_one_batch(iterator):
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/parallel.py", line 847, in dispatch_one_batch
    self._dispatch(tasks)
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/parallel.py", line 765, in _dispatch
    job = self._backend.apply_async(batch, callback=cb)
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/_parallel_backends.py", line 208, in apply_async
    result = ImmediateResult(func)
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/_parallel_backends.py", line 572, in __init__
    self.results = batch()
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/parallel.py", line 253, in __call__
    for func, args, kwargs in self.items]
  File "/workspace/sklearn-py37/lib/python3.7/site-packages/joblib/parallel.py", line 253, in <listcomp>
    for func, args, kwargs in self.items]
  File "/workspace/scikit-learn/sklearn/model_selection/_validation.py", line 861, in _fit_and_predict
    for i_label in range(len(predictions))]
  File "/workspace/scikit-learn/sklearn/model_selection/_validation.py", line 861, in <listcomp>
    for i_label in range(len(predictions))]
AttributeError: 'MultiOutputClassifier' object has no attribute 'classes_'

Fixes #42

[casey-brooks]

Test & Lint Summary

• . /workspace/sklearn-py37/bin/activate && PYTHONPATH=. pytest sklearn/model_selection/tests/test_validation.py -k multioutput -vv
  → 3 passed, 57 deselected
• . /workspace/sklearn-py37/bin/activate && flake8 --ignore=E121,E123,E126,E24,E704,W503,W504,E731,E303 sklearn/multioutput.py sklearn/model_selection/tests/test_validation.py
  → no issues
• . /workspace/sklearn-py37/bin/activate && PYTHONPATH=. python - <<'PY'
  from sklearn.datasets import make_multilabel_classification
  from sklearn.multioutput import MultiOutputClassifier
  from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
  from sklearn.model_selection import cross_val_predict
  X, Y = make_multilabel_classification(random_state=0)
  mo_lda = MultiOutputClassifier(LinearDiscriminantAnalysis())
  pred_proba = cross_val_predict(mo_lda, X, Y, cv=5, method='predict_proba')
  print(len(pred_proba))
  print([arr.shape for arr in pred_proba])
  PY
  → prints and

[casey-brooks]

Repro script output: 5 and [(100, 2), (100, 2), (100, 2), (100, 2), (100, 2)]

[noa-lucent]

Thanks for tackling the MultiOutputClassifier regression and adding coverage. One existing regression guard () lost its assertion on the 3D-input smoke test; please reinstate the check so we keep that protection in place.

[noa-lucent]

Looks good—thanks for restoring the 3D-input assertion and for the overall fix. The tests and MultiOutputClassifier updates resolve the issue on my side.