Paul Koch

Hi @LennartPurucker -- I looked at the updated code and I have two further recommendations: If the type hints to the _fit function are correct, then X and y are...

I had a further question on the n_jobs parameter. The current EBM implementation parallelizes the creation of the outer bags. Our default is to create 14 outer bags, so on...

To answer some of the questions above that I haven't answered yet: I agree adding a timeout parameter would be useful, and especially in the context of AutoGluon. I'll have...

> AutoGluon would parallelize the outer bags using Ray. For larger datasets, however, AutoGluon might give the model multiple cores; ideally, the model should still be able to use multiple...

> For now that remains an open question. We support feature importance and SHAP as black-box methods, but we don't have a clear vision for what interpretability in an AutoML...

Very nice. Out of curiosity, if it's easy to retrieve, what was the improvement observed on the preliminary tests, and what version of interpret was tested? After a basic version...

@Innixma -- Your implementation looked good to me. Has it been running successfully? I'd like to get the basics working first, but could add extras like multiclass pairs, etc, if...

Hi @MMShamszadeh -- Yes, we use newton boosting by default. The C API supports gradient boosting in addition as an option. In the C API this is controlled by the...

Hi @mersoz-rccl -- I can see why you'd want to do this for linear regression. I could be wrong, but my guess is that it wouldn't have a measurable benefit...

@mersoz-rccl - That's an interesting case. I think you're right that in many cases a linear model would perform well on such a dataset. Have you compared an unrestricted EBM...