lrp for regression

[bmayaci]

Dear @KaiFabi ,

Thanks for this implementation. I think it is super cool! Can your implementation be used for the regression problem where there is only one output neuron representing a numerical value. How should I change the implementation to do this, if possible?

Looking forward to hearing your comment on this!

Thanks

[kaifishr]

Hi @bmayaci

I have never tried relevance propagation for a regression task. At the moment, however, I don't see why it shouldn't work.

In lrp.py, line 104, you probably don't need the softmax function when working on a regression problem:

relevance = torch.softmax(activations.pop(0), dim=-1) # Unsupervised

In this case relevance = activations.pop(0) should be enough.