use highway network and employ decaying learning rate #58

primetang · 2016-12-28T05:00:14Z

use highway network and employ decaying learning rate in the new code.

dennybritz · 2016-12-28T22:17:53Z

train.py

@@ -126,15 +134,21 @@
        checkpoint_prefix = os.path.join(checkpoint_dir, "model")
        if not os.path.exists(checkpoint_dir):
            os.makedirs(checkpoint_dir)
-        saver = tf.train.Saver(tf.global_variables())
+        try:
+            saver = tf.train.Saver(tf.global_variables())


Why is the try/catch needed? Can't you just say saver = tf.train.Saver()?

dennybritz · 2016-12-28T22:23:55Z

Also, I think batches_num_pre_epoch is somewhat of a strange name. I think it should be num_batches_per_epoch. I also think it would be cleaner to not decay per epoch, but decay per some number of steps and have a parameter for that and get rid of that batches per epoch function.

primetang · 2016-12-29T01:19:58Z

Thank you for your attention, I have rewrite the code about decay_step, and use saver = tf.train.Saver() instead of try/catch.

qwaider · 2017-01-31T19:45:12Z

Interested

primetang added 2 commits December 28, 2016 12:54

use highway network and employ decaying learning rate

223b0a9

use highway network and employ decaying learning rate

12386a8

dennybritz reviewed Dec 28, 2016

View reviewed changes

rewrite the code about decay_step

93cbdf4

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

use highway network and employ decaying learning rate #58

use highway network and employ decaying learning rate #58

primetang commented Dec 28, 2016

dennybritz Dec 28, 2016 •

edited

Loading

dennybritz commented Dec 28, 2016

primetang commented Dec 29, 2016

qwaider commented Jan 31, 2017

use highway network and employ decaying learning rate #58

Are you sure you want to change the base?

use highway network and employ decaying learning rate #58

Conversation

primetang commented Dec 28, 2016

dennybritz Dec 28, 2016 • edited Loading

Choose a reason for hiding this comment

dennybritz commented Dec 28, 2016

primetang commented Dec 29, 2016

qwaider commented Jan 31, 2017

dennybritz Dec 28, 2016 •

edited

Loading