double counting of a patch?

[chail]

This was a really interesting paper! Thanks for providing the code.

I had adapted parts of this pipeline to work with a custom TF model. I was finding that in prediction_difference_analysis.py one of the patches was evaluated twice (by visualising the counts variable, and adding a check that every window = windows[w] is unique). I believe it occurs in lines 140-141, in which the upper index is win_idx+1, but win_idx is already incremented in line 109?

Another minor comment, should line 217 check that self.prob_tar=True?

Thank you!