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

sklearn/model_selection/tests/test_validation.py

@@ -60,6 +60,8 @@
 from sklearn.neighbors import KNeighborsClassifier
 from sklearn.svm import SVC
 from sklearn.cluster import KMeans
+from sklearn.multioutput import MultiOutputClassifier
+from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
 
 from sklearn.impute import SimpleImputer
 
@@ -936,6 +938,56 @@ def test_cross_val_predict_input_types():
     assert_array_equal(predictions.shape, (150,))
 
 
+def test_cross_val_predict_multioutput_predict_proba_list():
+    X, y = make_multilabel_classification(n_samples=120, n_features=10,
+                                          n_classes=4, random_state=0)
+    estimator = MultiOutputClassifier(LinearDiscriminantAnalysis())
+    proba = cross_val_predict(estimator, X, y, cv=3, method='predict_proba')
+
+    assert isinstance(proba, list)
+    assert len(proba) == y.shape[1]
+
+    expected_row_sums = np.ones(X.shape[0])
+    for idx, output_proba in enumerate(proba):
+        n_classes = np.unique(y[:, idx]).shape[0]
+        assert output_proba.shape == (X.shape[0], n_classes)
+        assert_allclose(output_proba.sum(axis=1), expected_row_sums,
+                        atol=1e-7)
+
+
+def test_cross_val_predict_multioutput_predict_proba_mixed_classes():
+    X_bin, y_binary = make_classification(n_samples=150, n_features=5,
+                                          n_informative=5, n_redundant=0,
+                                          n_classes=2, random_state=0)
+    X_multi, y_multiclass = make_classification(
+        n_samples=150, n_features=3, n_informative=3, n_redundant=0,
+        n_classes=3, n_clusters_per_class=1, random_state=1)
+    X = np.hstack([X_bin, X_multi])
+    y = np.column_stack([y_binary, y_multiclass])
+
+    estimator = MultiOutputClassifier(
+        LogisticRegression(max_iter=2000, solver="lbfgs"))
+    proba = cross_val_predict(estimator, X, y, cv=3, method='predict_proba')
+
+    assert isinstance(proba, list)
+    assert len(proba) == 2
+    assert proba[0].shape == (X.shape[0], 2)
+    assert proba[1].shape == (X.shape[0], 3)
+
+    expected_row_sums = np.ones(X.shape[0])
+    assert_allclose(proba[0].sum(axis=1), expected_row_sums, atol=1e-7)
+    assert_allclose(proba[1].sum(axis=1), expected_row_sums, atol=1e-7)
+
+
+def test_cross_val_predict_multioutput_without_method():
+    X, y = make_multilabel_classification(n_samples=90, n_features=8,
+                                          n_classes=3, random_state=1)
+    estimator = MultiOutputClassifier(LinearDiscriminantAnalysis())
+    predictions = cross_val_predict(estimator, X, y, cv=3)
+
+    assert predictions.shape == y.shape
+
+
 @pytest.mark.filterwarnings('ignore: Using or importing the ABCs from')
 # python3.7 deprecation warnings in pandas via matplotlib :-/
 def test_cross_val_predict_pandas():

sklearn/multioutput.py

@@ -320,11 +320,28 @@ class MultiOutputClassifier(MultiOutputEstimator, ClassifierMixin):
     ----------
     estimators_ : list of ``n_output`` estimators
         Estimators used for predictions.
+    classes_ : list of arrays
+        Class labels for each output, of shape ``(n_outputs,)``. Each entry is
+        an array of shape ``(n_classes_i,)`` containing the class labels for
+        the corresponding output.
     """
 
     def __init__(self, estimator, n_jobs=None):
         super().__init__(estimator, n_jobs)
 
+    def fit(self, X, y, sample_weight=None):
+        super().fit(X, y, sample_weight=sample_weight)
+        self.classes_ = [est.classes_ for est in self.estimators_]
+        return self
+
+    @if_delegate_has_method('estimator')
+    def partial_fit(self, X, y, classes=None, sample_weight=None):
+        super().partial_fit(X, y, classes=classes, sample_weight=sample_weight)
+        if not all(hasattr(est, 'classes_') for est in self.estimators_):
+            return self
+        self.classes_ = [est.classes_ for est in self.estimators_]
+        return self
+
     def predict_proba(self, X):
         """Probability estimates.
         Returns prediction probabilities for each class of each output.
@@ -420,7 +437,7 @@ def fit(self, X, Y):
             if self.order_ == 'random':
                 self.order_ = random_state.permutation(Y.shape[1])
         elif sorted(self.order_) != list(range(Y.shape[1])):
-                raise ValueError("invalid order")
+            raise ValueError("invalid order")
 
         self.estimators_ = [clone(self.base_estimator)
                             for _ in range(Y.shape[1])]