Description

I'm implementing a binary classifier. To support multi-class problems, I copied the approach from Gaussian process classification: if n_classes > 2, wrap the binary classifier in a OneVsOneClassifier or OneVsRestClassifier.

However, this is incompatible with me adding additional arguments to fit(), because the wrappers don't forwards these arguments to the wrapped binary classifier.

AIUI the fix would be to change all fit(X, y) methods in sklearn.multiclass to fit(X, y, **kwargs), and pass that new argument to the wrapped.fit(X, y, **kwargs) call.

Should I simply create a merge request, or am I missing something?

Versions

Linux-4.18.5-arch1-1-ARCH-x86_64-with-arch
Python 3.7.0 (default, Jul 15 2018, 10:44:58) 
[GCC 8.1.1 20180531]
NumPy 1.15.0
SciPy 1.1.0
Scikit-Learn 0.19.2