scikit-learn-contrib/lightning

CDClassifer consistency

arthurmensch opened this issue · 1 comments

The API of CDClassifier and SAG(A)Classifier should be uniformized.

For the moment, the objective of CDClassifier is parametrized by C, loss alpha and penalty,
whereas SAG(A)Classifier sets C = 1 / n_samples, uses beta instead of alpha and a specific alpha to add a ridge regularization. Wouldn't it be better to align the CDClassifier API on the SAG(A)Classifier ?

+1 from me.