Fix shape function plotting.

[JulienSiems]

No description provided.

[JulienSiems]

@amueller I merged your changes and fixed the column labeling.

[amueller]

Hm it feels strange that we make up a bin on the left but not the right. Shouldn't we have to make up a bin on both sides? Maybe you need to make up a bin on the end and duplicate the last weight to get the correct plot? Cause otherwise the last step has a bin of width 0 associated with it?

[amueller]

an alternative might be to use bar instead of step, which would maybe yield a sightly less cumbersome computation, but I think only slightly. Can you maybe show a plot on a simple example?

[JulienSiems]

I understand that it isn't the cleanest fix, but I think for the paper we will want to use probably the shape function plotting from the EBM repo? Maybe this is sufficient for the plots we are doing right now?

[amueller]

@JulienSiems I don't think @Harsha-Nori will have time to do the integration with the EBM vis, I'm not sure if either of us will be able to do it?

[Harsha-Nori]

Hey! I'm unfortunately summoned on travel again, but I pushed some WIP code for shape function plotting using interpret's visualizations onto the develop branch (alongside an example notebook). Currently it only works on binary classification with numeric features, but now that I have things sketched out, I can extend support for categoricals, multiclass, and overall feature importance once I'm back in Seattle.