Use SHAP Values for PyTorch RNN / LSTM

Question

Is there a way to do the above? The SHAP Package is very helpful and works pretty well for PyTorch Neural Nets. For PyTorch RNNs i get the error message below (for LSTMs its the same):

Seems like it doesn't work but is there a workaround or something? Does anyone have experience with PyTorch and SHAP?

SvitlanaGA...supportsUkraine · Accepted Answer · 2020-01-27 13:13:27Z

5

'RNNs aren't yet supported for the PyTorch DeepExplainer (A warning pops up to let you know which modules aren't supported yet: Warning: unrecognized nn.Module: RNN). In this case, the explainer assumes the module is linear, and makes no change to the gradient. Since RNNs contain nonlinearities, this is probably contributing to the problem.' That was an answer I found at Shap.

Try to check captum.ai that is built on PyTorch.

answered Jan 27, 2020 at 13:13

SvitlanaGA...supportsUkraine

1,4192 gold badges18 silver badges24 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

tamnva Jan 20 at 18:33

I am having the same issue with shap=0.46.0. The think with captum.ai is that it does not have the something similar to shap.DeepExplainer

Collectives™ on Stack Overflow

Use SHAP Values for PyTorch RNN / LSTM

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related