FIX Make skorch work with sklearn 1.6.0

skorch/classifier.py

@@ -231,6 +231,12 @@ def predict(self, X):
         """
         return self.predict_proba(X).argmax(axis=1)
 
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'classifier'
+        return tags
+
 
 neural_net_binary_clf_doc_start = """NeuralNet for binary classification tasks
 
@@ -379,3 +385,9 @@ def predict(self, X):
         """
         y_proba = self.predict_proba(X)
         return (y_proba[:, 1] > self.threshold).astype('uint8')
+
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'classifier'
+        return tags

skorch/net.py

@@ -2725,3 +2725,11 @@ def __repr__(self):
 
         parts.append(')')
         return '\n'.join(parts)
+
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added for
+        # NeuralNet, NeuralNetClassifier, etc. See:
+        # https://scikit-learn.org/dev/developers/develop.html#estimator-tags
+        # https://scikit-learn.org/dev/modules/generated/sklearn.utils.Tags.html#sklearn.utils.Tags
+        tags = BaseEstimator.__sklearn_tags__(self)
+        return tags

skorch/probabilistic.py

@@ -657,6 +657,12 @@ def fit(self, X, y=None, **fit_params):
         self.partial_fit(X, y, **fit_params)
         return self
 
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'regressor'
+        return tags
+
 
 gp_regr_doc_start = """Gaussian Process regressor
 
@@ -729,6 +735,12 @@ def __init__(
             **kwargs
         )
 
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'regressor'
+        return tags
+
 
 gp_binary_clf_doc_start = """Gaussian Process binary classifier
 
@@ -912,6 +924,12 @@ def predict(self, X):
         y_proba = self.predict_proba(X)
         return (y_proba[:, 1] > self.threshold).astype('uint8')
 
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'classifier'
+        return tags
+
 
 # BB: I could never get any reasonable results using ``SoftmaxLikelihood``. In
 # fact, it always produces NaN. Probably I use it wrongly but there are no

skorch/regressor.py

@@ -80,3 +80,9 @@ def fit(self, X, y, **fit_params):
         # this is actually a pylint bug:
         # https://github.com/PyCQA/pylint/issues/1085
         return super(NeuralNetRegressor, self).fit(X, y, **fit_params)
+
+    def __sklearn_tags__(self):
+        # TODO: this is just the bare minimum, more tags should be added
+        tags = super().__sklearn_tags__()
+        tags.estimator_type = 'regressor'
+        return tags

skorch/tests/test_helper.py

@@ -437,6 +437,7 @@ def test_grid_search_with_slds_works(
             self, slds, y, classifier_module):
         from sklearn.model_selection import GridSearchCV
         from skorch import NeuralNetClassifier
+        from skorch.utils import to_numpy
 
         net = NeuralNetClassifier(
             classifier_module,
@@ -450,12 +451,13 @@ def test_grid_search_with_slds_works(
         gs = GridSearchCV(
             net, params, refit=False, cv=3, scoring='accuracy', error_score='raise'
         )
-        gs.fit(slds, y)  # does not raise
+        gs.fit(slds, to_numpy(y))  # does not raise
 
     def test_grid_search_with_slds_and_internal_split_works(
             self, slds, y, classifier_module):
         from sklearn.model_selection import GridSearchCV
         from skorch import NeuralNetClassifier
+        from skorch.utils import to_numpy
 
         net = NeuralNetClassifier(classifier_module)
         params = {
@@ -465,7 +467,7 @@ def test_grid_search_with_slds_and_internal_split_works(
         gs = GridSearchCV(
             net, params, refit=True, cv=3, scoring='accuracy', error_score='raise'
         )
-        gs.fit(slds, y)  # does not raise
+        gs.fit(slds, to_numpy(y))  # does not raise
 
     def test_grid_search_with_slds_X_and_slds_y(
             self, slds, slds_y, classifier_module):