Open bnaul opened 7 years ago
I am very much in favor of this proposal.
Me too, and would like to hear more about what you have in mind regarding point 2. :+1:
Additionnally, keeping api matching with Pandas and numpy would be a plus for better usage.
Currently we have a functional API with static methods like
build_model.build_model_from_featureset
andpredict.model_predictions
. Not only are these a bit of a mouthful, they're also a bit unintuitive because of how differently they are used than the correspondingsklearn
functionality. I suggest that we try to modify our API to be more like that ofsklearn
, specifically:1) Add a
Model
class that wraps asklearn
model and provides ansklearn
-like API (.fit
,.predict
,.predict_proba
) for interacting withFeatureset
objects. 2) (optional) Also changefeaturize
to behave more like ansklearn
preprocessor; maybe you'd initialize aFeatureset
and call.transform
on a dataset? This is less obvious but there's probably some improvement to be made here as well.