Multiclass classification fast shap

[valevalerio]

Hello,
thanks for sharing the runnable code from your paper, looking forward to include fastshap among my fav explainers!

I would like to know if there is a stright forward adaptation of FS to a multiclass classification (i.e. more than 2 classes).
I have tried to change directly the number of inputs neurons and output neurons for the surrogate model. From the tutorial, the surrogate model has 2 * num_features inputs and 2 outputs. I changed it in order to handle n_classes * num_features. And similarly in in the explainer model, i changed it to have n_classes * num_features outputs rather than just 2 * num_features.

Maybe I am oversimplifying, but nontheless seems a smooth change, in the execution fails.
Am I overlooking for a solution?

[iancovert]

Hi Valerio, thanks for checking out the repo!

FastSHAP is easy to adapt to models with >2 classes. If you have $k$ classes then the original model and surrogate will both have $k$ outputs. On the input side, if you have $d$ features then the original model will have $d$ inputs, and the surrogate will have $2d$ (because we pass a binary indicator of which features are missing). Finally, the explainer will have $d$ inputs and $dk$ outputs, with one Shapley value per feature for each class.

The difference with what you've described seems to be with the surrogate: the surrogate should have $2d$ inputs and $k$ outputs.

Let me know if this makes sense, hopefully the modified code works now!

[valevalerio]

Thank you very much for the tempestivirty of your reply.
The indicators of missing features in the surrogate model were indeed the catch.
It worked as a C H A R M.
Thanks