Exception in KernelExplainer.solve() about shap HOT 4 CLOSED

It looks like it might have selected an all-zero explanation and then failed. I'll look into it and try and reproduce the problem. Thanks for letting me know!

from shap.

Thanks. Can you give me some clues that I can try to fix the problem?

-Param

from shap.

I just pushed changes that should fix the problem. It showed up when the feature selection phase found no important features. So the fix just returns a zero explanation in that case.

You can also turn off feature selection by passing KernelExplainer the param use_l1=False, then the regression estimation will assume all features have non-zero impact and not do debiasing feature selection. You can also increase the number of samples with the nsamples param.

from shap.

Closing this issue unless I hear otherwise.

from shap.