Breaking change in scikit-Learn for ClassifierMixing and BaseEstimator

[jverce]

Newer versions of scikit-learn (v1.6.0 onwards) changed how they detect the type of an estimator (i.e. classifier vs. regressor). Previous versions relied on the _estimator_type attribute of the estimator. Newer versions however read the estimator_type tag instead, which is a tag that's defined both by the BaseEstimator and the ClassifierMixin classes (see the actual code change for more details).

So the following class (from chapter 7 of the Machine Learning with PyTorch and Scikit-Learn book) will start throwing warnings:

class MajorityVoteClassifier(BaseEstimator, ClassifierMixin):
  # ...

Warnings will look like this:

ValueError: MajorityVoteClassifier should either be a classifier to be used with response_method=predict_proba or the response_method should be 'predict'. Got a regressor with response_method=predict_proba instead.

The reason is that the tag values are based on whatever BaseEstimator provides (which in the estimator_type case, it's just None). To make the example work, you can just reverse the order of the parent classes of MajorityVoteClassifier, so that the tags provided by ClassifierMixin take precedence:

-class MajorityVoteClassifier(BaseEstimator, ClassifierMixin):
+class MajorityVoteClassifier(ClassifierMixin, BaseEstimator):

[rasbt]

I had no idea and really appreciate the note! Just addressed it here: #206.
Many thanks!