enhancement base class for sklearn models

Describe the workflow you want to enable

Sklearn does not provide an abstract class for the fit method and transform method which is not friendly to code hints. In Python project, type hints is very important. This would be helpful for python type hints if sklearn had abstract classes with fit and transform methods

=================================== i hope a abstract class like this

class IFit:
    def fit(x, y, **params):
        pass

class ITransform:
    def transform(x, y, **params):
        pass

Describe your proposed solution

class IFit(meteclass=abc.ABCMeta):
    def fit(x, y, **params):
        pass

class ITransform(meteclass=abc.ABCMeta):
    def transform(x, y, **params):
        pass

class BaseClassifier(IFit, ITransform, BaseEstimator, ClassifierMixin):
    pass

class BaseRegressor(IFit, ITransform, BaseEstimator, RegressorMixin):
    pass

class BaseCluster(IFit, ITransform, BaseEstimator, ClusterMixin):
    pass

Describe alternatives you’ve considered, if relevant

No response

Additional context

No response

[thomasjpfan]

typing’s Protocol was designed to allow for duck typing. The Python PEP that describes the Protocol has examples that promotes duck typing: https://www.python.org/dev/peps/pep-0544/. One can say a Protocol “explicitly defines an interface”.

The way scikit-learn estimators are duck typed prevents us from having type checking using a class hierarchy. We do not require third party estimators to inherit our classes, such as BaseEstimator. As long as the third party estimator defines the required methods, it should work with sklearn’s functions/meta-estimators. In other words, if we have abstract classes, we can not use them typing in sklearn, because we do not require third party estimators to inherit from them.

[franklucky001]

@thomasjpfan Implicit implementation interface is not pythonic, fit & transform are not standard protocals of python.I know that adding these features requires major changes in the code, this is just a suggestion.