search

johnmwu / contextual-corr-analysis

17 stars 5 forks source link

Structure of activations #1

Closed boknilev closed 5 years ago

boknilev commented 5 years ago

I think it's possible that activations.dim() == 1. The new commit excludes this option: https://github.com/johnmwu/contextual-corr-analysis/commit/1a57e51260915b5b9978d683392831c35012f6f8#diff-6956be0ba493db843ef9076f20ff4e3dR42, which existed before. Please double-check.

johnmwu commented 5 years ago

What would activations.dim() == 1 mean? I was under the impression that activations was either a shape (nlayer, len(sent), nneuron) or (len(sent), nneuron) tensor.

Because I don't know what this would even mean, my code doesn't handle this case (simply removing the exception may lead to surprises).

boknilev commented 5 years ago

Makes sense.