added hack-support for n_features_to_select

scikit-learn-contrib / boruta_py

Python implementations of the Boruta all-relevant feature selection method.

BSD 3-Clause "New" or "Revised" License

1.46k stars 252 forks source link

added hack-support for n_features_to_select #93

Closed bgalvao closed 3 years ago

bgalvao commented 3 years ago

allows user to specify a number of features to select
this depends on the feature importance of the last iteration's ensemble fit (of accepted + tentative features)
features that were rejected are left out from this selection
pull request is just a proposal to issue #92

danielhomola commented 3 years ago

Thanks a lot for the PR, really appreciate the effort, but I honestly don't see the point of this.

Getting what you want is literally one line (as you found out too: feature_importance.argsort()[-n:]

and it's a concern of downstream analysis after feature selection, nothing to do with Boruta.

It's as if you added a method to 'PCA(n_components=25)' that subsets the components and singular values to the first n ones...

bgalvao commented 3 years ago

It is just like you described in the PCA example. Fair enough, it was just for the sake of added functionality, and I can see why it beats the original purpose of Boruta.