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

Removed redundant tf.exp on predictions in evaluate method

Browse Source 
In the `evaluate` method, the usage of `tf.exp` seems redundant here, since `predictions` itself is the logits output of a dense layer in decoder. Therefore, I removed `tf.exp` so that `tf.multinomial` is applied on the logits directly, which agrees better with the proper usage of `tf.multinomial`.
This commit is contained in:
Ruizhi 2018-08-02 14:16:04 +08:00 committed by GitHub
parent 842cd17c10
commit f0b3a02cb7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb
View File

@ -753,7 +753,7 @@
" attention_weights = tf.reshape(attention_weights, (-1, ))\n",
" attention_plot[t] = attention_weights.numpy()\n",
"\n",
" predicted_id = tf.multinomial(tf.exp(predictions), num_samples=1)[0][0].numpy()\n",
" predicted_id = tf.multinomial(predictions, num_samples=1)[0][0].numpy()\n",
"\n",
" result += targ_lang.idx2word[predicted_id] + ' '\n",
"\n",