Supporting EBMs (Explainable Boosting Machines)

[seb-mueller]

Hi, I was wondering if there are plans to support EBMs since they appear quite promising in terms of performance (matching XGB etc) whilst remaining very interpretable.

There is even an R package interpret, but I haven't found a way to make it play nicely with parsnip.

[simonpcouch]

Thanks for the issue, @seb-mueller. This sounds like it may be a good fit for the bonsai parsnip extension package, so I will go ahead and move this issue there. :)

[frankiethull]

@seb-mueller, @simonpcouch, @EmilHvitfeldt -

I have some interesting-ish news.

I created an experimental package called daisugi that registers EBMs to parsnip. I have it registered as a boost_tree & a unique new model type called explainable_boost, since EBMs are kind of unique in their own way.

Feel free to test EBMs out in daisugi but note daisugi is not on CRAN.

tidymodels work-in-progress documentation, daisugi is curating weird tree models & am focused on models not on CRAN at the moment. I don't plan on registering all of them to parsnip, buttttt wanted to test out EBMs myself in an MLOps pipeline so had to register it 😄