erikyao / mysemiboost Goto Github PK

Enable API users to select scoring of their interest.

May need to change accum_decision_function, accum_predict and accum_predict_proba.

You can return a list of classifiers of [SB(t=1), SB(t=2), ..., SB(t=20)] where each SB(t=i) is an partial estimator from SB.

Limited scoring parameter to be an dictionary of {string, scorer_object}. See _multimetric_score for how to use scorer objects.

This refactoring will also solve the issue of some classifiers not implementing decision_functions.