Features importance for Hist Gradient Boosting

M3. Hyperparameter tuning
#1

Hi :slight_smile:

Since I did not get an answer to my question regarding if features importance is supported for Hist Gradient Boosting Classifier (say for one of the exercise in module 3). I wonder if I should get it as no?! and if so, do we have a work around like PCA or some experimental version of Hist Gradient Boosting that supports features importance?

Best wishes for the new year and thanks in advance,
Tsvika

#2

I answered the previous question. permutation_importance is the way to go. PCA is not because it make represent a linear combination of the original feature and not the original features themselves :slight_smile:

1 Like