We want to improve HSTree by selecting the regularization parameter using the GCV ridge formula.
This requires extracting tree features and then just using sklearn's ridge class to compute GCV for different regularization parameters. @aagarwal1996 can you implement this by using the gMDI code to do the extraction?
For ensembles, we also want parallelize over trees (see #125) and maybe use out-of-bag samples for trees to perform cross-validation.
In #174, @OmerRonen sped up the HSTree fitting by changing the code to vary only the regularization parameter rather than refit many trees.
We want to improve HSTree by selecting the regularization parameter using the GCV ridge formula.
This requires extracting tree features and then just using sklearn's ridge class to compute GCV for different regularization parameters. @aagarwal1996 can you implement this by using the gMDI code to do the extraction?
For ensembles, we also want parallelize over trees (see #125) and maybe use out-of-bag samples for trees to perform cross-validation.
In #174, @OmerRonen sped up the HSTree fitting by changing the code to vary only the regularization parameter rather than refit many trees.
cc @Innixma