search

suinleelab / path_explain

A repository for explaining feature attributions and feature interactions in deep neural networks.
MIT License
185 stars 28 forks source link

Make num_samples behave same for torch setup #11

Closed jumelet closed 3 years ago

jumelet commented 3 years ago

As discussed in issue #8, the torch setup with use_expectations=False now behaves the same as for Tensorflow models.

Btw, I had to add num_samples-1 in the fraction as well, because the range(0, num_samples) now ranges from 0 to num_samples-1. I checked the output of the two model types and the output is now the same.