Enabled feature_importances_ for our ForestDML and ForestDRLearner estimators#306

Merged

vsyrgkanis merged 17 commits intomasterfrom vasilis/feature_importances

Nov 9, 2020

+399-268

Commits on Nov 7, 2020

Commits on Nov 8, 2020

Merge branch 'master' into vasilis/feature_importances
vsyrgkanis
authored
Update econml/sklearn_extensions/ensemble.py

vsyrgkanis
and
kbattocchi
authored
changed order of mixins
vasilismsr
committed
Merge branch 'vasilis/feature_importances' of github.com:microsoft/EconML into vasilis/feature_importances
vasilismsr
committed
changed feature_improtances_ implementation as normalization at the tree level was causing trouble, since due to sample splitting feature_improtance can many times be negative (increase in variance…
vasilismsr
committed
fixed docstring reference
vasilismsr
committed
fixed the problem with inconsistent impurities in subsampled honest forest, that now makes feature_importances_ exactly correct and no need to re-implement the method. Now impurities are computed o…
vasilismsr
committed
fixed docstring doctest
vasilismsr
committed

Commits on Nov 9, 2020