Michael Mayer
Michael Mayer
Ahh, actually, I once had a look at these functions in c_api.cpp! We can add these and a `num_iteration()` method to the Booster class and use that to make clean...
Pipeline seems happy @jameslamb - but really no pressure :-)
> Thanks very much for this! And I really appreciate you keeping it up to date with `master` while waiting for reviews. > > I just left one suggestion about...
> Thanks so much! > > Thanks for working with me on the unit tests. I'm really happy that this PR is going to add so much better coverage of...
> yes please done.
@jameslamb : Thanks! In "shapviz", that uses LGB as minimal "Enhances" dependency, I am using the snippet in https://github.com/mayer79/shapviz/blob/main/R/shapviz.R ```r if (!requireNamespace("lightgbm", quietly = TRUE)) { stop("Package 'lightgbm' not installed")...
It seems obvious that the importance is calculated on the features *before preprocessing*. Especially with permutation importance, permuting dummy columns is non-sensical as suddenly two dummies would be 1 within...
Maybe using {ggside}? (Not sure if it does what you have in mind though)
Please add a minimal reproducible example.
@nlebovits I am not maintainer. SHAP can't be calculated for non-probabilistic classification models. It requires numeric predictions, e.g., logit probabilities. There is an open PR that allows probabilistic classification for...