Add fit_cate_intercept to DML, rework feature generation #174
Conversation
|
Fixes #119. |
kbattocchi
force-pushed
the
kebatt/dmlIntercept
branch
from
8c28f52 to
8497b06
Compare
There was a problem hiding this comment.
There is a weirdness now if I want to get the feature names. There is no good way to access the original featurizer, as the original featurizer is embedded in the new pipeline. For instance I cannot do:
est.featurizer.get_feature_names(['A', 'B', 'C', 'D'])
However, the featurizer method is the right thing, as this is what the inference methods need to access.
I think the best would be to add the method: cate_feature_names, as I had done in the DRLearner, which will access the second stage of the pipeline and call the get_feature_names. This will uniformize even more the API between the two methods.
…docstrings to these and to the models_y and models_t methods so that they appear on the module docs. This uniformizes the interface with the DRLearner.
… to (n_y, n_t, n_x). Fixed corresponding tests. Reverted cross_product back
…EconML into kebatt/dmlIntercept
There was a problem hiding this comment.
@vasilismsr
As the original PR author, I can't request changes, but please address these comments.
Co-Authored-By: Keith Battocchi <[email protected]>
…EconML into kebatt/dmlIntercept
kbattocchi
dismissed
vasilismsr’s stale review
Dismissing review since changes have been addressed and we're trying to get out the release.
No description provided.