search

stanfordmlgroup / ngboost

Natural Gradient Boosting for Probabilistic Prediction
Apache License 2.0
1.64k stars 215 forks source link

SHAP doesn't work if LogNormal #186

Open pkdmark opened 3 years ago

pkdmark commented 3 years ago

Hello,

Problem can be reproduced with a slight modification to the notebook below:

https://github.com/stanfordmlgroup/ngboost/blob/master/examples/model_interpretation/SHAP_plots.ipynb

If the distribution is LogNormal (rather than normal) you will find that the Base Value + Shap Values of a single observation do not equal the prediction.

Penna88 commented 11 months ago

I had a similar issue with Gamma dist and I commented as #291 Hope it helps.