Fix issues with sklearn 1.4

DON'T MERGE YET.

Some skorch tests fail with sklearn v1.4. This commit fixes them:

1. Inheritance structure of scorers seems to have changed.
2. classes_ attribute should always be numpy array
3. CalibratedClassifierCV only works with predict_proba being float64

To fix the latter, I'm now casting the output of predict_proba to
float64. However, I'm not sure if this is a good idea, as it might break
existing code. I'm just adding it in for now to check if the tests pass.

skorch/callbacks/scoring.py

@@ -100,7 +100,7 @@ def convert_sklearn_metric_function(scoring):
 
         # those are scoring objects returned by make_scorer starting
         # from sklearn 0.22
-        scorer_names = ('_PredictScorer', '_ProbaScorer', '_ThresholdScorer')
+        scorer_names = ('_PredictScorer', '_ProbaScorer', '_ThresholdScorer', '_Scorer')
         if (
                 hasattr(module, 'startswith') and
                 module.startswith('sklearn.metrics.') and

skorch/classifier.py

@@ -97,7 +97,7 @@ def classes_(self):
             if not len(self.classes):
                 raise AttributeError("{} has no attribute 'classes_'".format(
                     self.__class__.__name__))
-            return self.classes
+            return np.asarray(self.classes)
 
         try:
             return self.classes_inferred_
@@ -301,7 +301,7 @@ def _default_callbacks(self):
 
     @property
     def classes_(self):
-        return [0, 1]
+        return np.array([0, 1])
 
     # pylint: disable=signature-differs
     def check_data(self, X, y):

skorch/net.py

@@ -1597,7 +1597,7 @@ def predict_proba(self, X):
             yp = nonlin(yp)
             y_probas.append(to_numpy(yp))
         y_proba = np.concatenate(y_probas, 0)
-        return y_proba
+        return y_proba.astype(np.float64)
 
     def predict(self, X):
         """Where applicable, return class labels for samples in X.