Closed ThomasBury closed 3 years ago
Sorry, completely forgot about this.
I applaud your effort, but I'm worried about a couple of things here, maybe you can shine some light on these:
As I see in your other comment, you've
No worries (a part of your post is missing, but here are some thoughts)
If you want so, you can choose from the Boruta-Shap package or some of the material of my PR (or not at all, but then we'll be left with different implementations of the globally same algorithm).
Hi,
Actually, the Shapley and permutation importances must be computed on unseen data, which is the case in my PR, not in Boruta_shap (I raised an issue on the boruta_shap github repo). KR
Hi,
Ekeany just wrote a package (https://github.com/Ekeany/Boruta-Shap) with the same modifications that I made in the PR (https://github.com/scikit-learn-contrib/boruta_py/pull/77).
To avoid duplicates and for the benefit of the community, it makes sense to merge those contributions. For the sake of clarity, under the scikit-learn-contrib umbrella.
Thanks! KR