Closed kodonnell closed 4 years ago
Hey @kodonnell, thanks for your contribution!
The reason I use split_value
is because the library supports categorical variables, whereas scikit-learn does not. This code is based on https://github.com/scikit-learn/scikit-learn/pull/12866.
I agree threshold should return something more sensible, but casting to a float will not show the right value for categorical variables.
I suggest returning a structure instead, that when casting to a string will yield either the threshold, or the indexes of categories.
Superseded by 3214b291410acbb5e0abf44dbd289670ea674caa
Fixes #12. Note that the internal use of
split_value
should probably be removed completely as per sklearn, but that can be a separate PR.