Very-Deep-Convolutional-Networks-for-Natural-Language-Processing-in-tensorflow icon indicating copy to clipboard operation
Very-Deep-Convolutional-Networks-for-Natural-Language-Processing-in-tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon geduo15

Incorrect batch_norm usage

Open shatu opened this issue 7 years ago • 0 comments

  1. The trainable argument in the batch_norm op should be replaced with is_trainable.
  2. I think the starting convolution layer is incorrect. In my opinion, this stride should be [1, 1, ,1, 1] instead of [1, 1, embedding_size, 1], and padding=Valid instead of padding=Same should be used.

shatu avatar Sep 25 '17 01:09 shatu