smazzanti / mrmr

mRMR (minimum-Redundancy-Maximum-Relevance) for automatic feature selection at scale.
MIT License
531 stars 79 forks source link

Allow "fixed" features #44

Open dannykspl opened 6 months ago

dannykspl commented 6 months ago

Sometimes a user may wish to require certain features to be present in the feature space (for inference purposes/interpretability, etc.) and determine what other features minimize redundancy and maximize relevance with those features present.

Suggestion: revise mrmr_regression/classif function to take an optional list of k "fixed_features" such that the function returns a list of N features that always includes those k features and N-k other features that minimize redundancy/maximize relevance in the presence of those k fixed features.