experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix shapes in comments of nmt_with_attention.ipynb

Browse Source 
It is a bit misleading and confusing that the output shape of decoder is currently commented as `(batch_size * max_length, vocab)`. However the correct shape should be `(batch_size * 1, vocab)`, since the input x of GRU layer has shape == `(batch_size, 1, embedding_dim + hidden_size)`.
This commit is contained in:
Ruizhi 2018-08-01 16:55:33 +08:00 committed by GitHub
parent abd645085b
commit 842cd17c10
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb
View File

@ -552,10 +552,10 @@
" # passing the concatenated vector to the GRU\n", " # passing the concatenated vector to the GRU\n",
" output, state = self.gru(x)\n", " output, state = self.gru(x)\n",
" \n", " \n",
" # output shape == (batch_size * max_length, hidden_size)\n", " # output shape == (batch_size * 1, hidden_size)\n",
" output = tf.reshape(output, (-1, output.shape[2]))\n", " output = tf.reshape(output, (-1, output.shape[2]))\n",
" \n", " \n",
" # output shape == (batch_size * max_length, vocab)\n", " # output shape == (batch_size * 1, vocab)\n",
" x = self.fc(output)\n", " x = self.fc(output)\n",
" \n", " \n",
" return x, state, attention_weights\n", " return x, state, attention_weights\n",