BUG: DeepExplainer explanations do not sum up to model output about shap

[CloseChoice]

Thanks for the report tn84. Is it possible for you to provide the model defintion and a dataset that reproduces this issue?

This is another issue with the deepexplainer that somehow does breaks additivity. Do we know more about that @connortann? I've read this now multiple times and also encountered it in my PR to fix LSTMs. Could we maybe get a DeepExplainer label?

[connortann]

Sure, I created a deep explainer label. Yep, to investigate further we really need a MRE: please see writing helpful bug reports.

[CloseChoice]

Sure, I created a deep explainer label. Yep, to investigate further we really need a MRE: please see writing helpful bug reports.

Thanks a lot! I got an example in my LSTM PR but need some more time to investigate this further and also a couple more examples might come in handy since there could be something wrong with a couple operation types so that we might assign wrong custom gradients to these.