Run with pre-computed SHAP values/interactions

Hello,

Getting familiar with your great tool, I was wondering how/if I could prevent computing SHAP values. I have already computed and saved SHAP values and interaction values on a GPU cluster and was wondering if there was a way to feed them to an Explainer and bypass their calculation when I run the explainer dashboard on my laptop.

Thanks a lot

[oegedijk]

Hey, forgot to loop back on this, but actually added methods explainer.set_shap_values() and explainer.set_shap_interaction_values() that allow you to import pre-calculated shap values in the last release.

Could you test if these work for you?

[e5k]

Hey @oegedijk ,

After doing a few tests, the SHAP values entered manually don’t make much sense. I calculate the SHAP values following this example and then input these shape values in the explainer. Attached are two pictures, the first showing SHAP values computed by the explainer, the second showing pre-computed Shap values entered with your suggestion.

Can you make any sense of that? I am messing up things?

Alternatively, is there any way to activate SHAP’s GPUTree explainer?

Thanks a lot