search

yandex / faster-rnnlm

Faster Recurrent Neural Network Language Modeling Toolkit with Noise Contrastive Estimation and Hierarchical Softmax
Other
561 stars 138 forks source link

Returning the hidden layers of the RNN #15

Closed tokestermw closed 8 years ago

tokestermw commented 8 years ago

Hi, thanks for the great package.

I am curious if it is possible to output the unrolled hidden layers of the RNN for each sentence? Specifically, for each new sentence, RNNLM can calculate a score (perplexity / log prob) and if it can also output the values for the hidden layer for each word, that would create a nice visualization.

tokestermw commented 8 years ago

I believe would be useful. https://github.com/yandex/faster-rnnlm/blob/master/faster-rnnlm/rnnlm.cc#L199