CODEWITHSUNDEEP
pythontensorflow
Sufeng Niu
Sufeng Niu

Reputation: 639

tensorflow sequence to sequence without softmax

I was using Tensorflow sequence to sequence example code. for some reason, I don't want to add softmax to output. instead, I want to get the raw output of decoder without softmax. I was wondering if anyone know how to do it based on sequence to sequence example code? Or I need to create it from scratch or modify the the seq2seq.py (under the /tensorflow/tensorflow/python/ops/seq2seq.py)?

Thank you

Upvotes: 0

Views: 394

Answers (1)

Benoit Steiner
Benoit Steiner

Reputation: 1469

The model_with_buckets() function in seq2seq.py returns 2 tensors: the output and the losses. The outputs variable contains the raw output of the decoder that you're looking for (that would normally be fed to the softmax).

Upvotes: 1

Related Questions