Option to specify features to use for splitting and for leaf models

cerlymarco / linear-tree

A python library to build Model Trees with Linear Models at the leaves.

MIT License

338 stars 54 forks source link

Option to specify features to use for splitting and for leaf models #3

Closed JonasRauch closed 2 years ago

JonasRauch commented 2 years ago

Added two additional parameters:

split_features: Indices of features that can be used for splitting. Default all.
linear_features: Indices of features that are used by the linear models in the leaves. Default all except for categorical features

This implements a feature requested in https://github.com/cerlymarco/linear-tree/issues/2

Potential performance improvement: Currently the code still computes bins for all features and not only for those used for splitting.

cerlymarco commented 2 years ago

Hi Jonas,

Thanks for your patience to look into my messy code, I really appreciate your valuable contribution.

At a first glance, it seems ok but I want to test it in detail. In the next months, I plan to make some extensions to the library with other functionalities. So I think, I'll merge your contribution in the next version.

Stay tuned

JonasRauch commented 2 years ago

I agree that this needs additional testing. In fact, it would be great to add some unit tests to the library that cover the most important use cases. Looking forward to the next version.

cerlymarco commented 2 years ago

I've merged it... Soon I'll release the new version

Thanks