Tobias Pitters

> @CloseChoice I might have closed #661 by accident, apologies to @riiswa if that was the case. > > If we can add what @axb2035 discussed in this issue as...

As it says there is currently no implementation for this. You can encode the categorical columns and then calculate the shap values.

The problem your having is likely due to having tensorflow 2.16 and/or keras 3 installed. This is just supported on master, not on the latest release

I don't think we have the capacity to implement this anytime soon

This is a longstanding problem we have with LSTMs. Currently there is no solution in sight for this

There is a problems with lstms in general, NLP or not. But I am not aware that we have problem with NLP use cases, could you please provide the code...

Could you please provide a minimal reproducible example, so sharing also how to create the MLP? And could you please open another issue for that, since it is not exactly...

Thanks for bringing this back up. I would agree that this is kind of important but not sure how much time I have to dig into this since a fix...

Thanks for the report and for your interest to contribute to shap. The warning is thrown [here](https://github.com/shap/shap/blob/master/shap/explainers/_deep/deep_pytorch.py#L242). In order to support the layers you point out, one needs to add...

@JulioHC00 something seems off there, the accepted threshold is way too high, should be something like `0.01`