Calculate test loss and perplexity for language model

Question

Calculate test loss and perplexity for language model

NirantK opened this issue 6 years ago · 3 comments

Answer 1 · 2018-10-04T10:29:51.000Z

@NirantK could you tell me how you calculated the perplexity for your model?
is it e^ loss ?

Answer 2 · 2018-10-04T10:38:06.000Z

The training (and/or validation) loss and perplexity are calculated during model training as shown in the notebook. This is done by fastai implementation of metrics via hooks.

We can repeat the same on test data. This is yet to be done - as indicated by this issue.

Answer 3 · 2018-10-04T10:40:09.000Z

You aren't printing perplexity anywhere, you've just mentioned it in a comment. So I had a doubt about the same. Thank you for your answer.