"git@developer.sourcefind.cn:OpenDAS/torch-scatter.git" did not exist on "a4355beaab4b85573a116caf975897ab02a80d22"
Commit dc340239 authored by Michael Figurnov's avatar Michael Figurnov
Browse files

Do not perform averaging of the cost.

The moving average object was shared between training and evaluation. As
a result, the cost during evaluation was averaged with the training
cost. This affects only objective, not precision.
parent a5c4fd06
...@@ -122,10 +122,7 @@ class ResNet(object): ...@@ -122,10 +122,7 @@ class ResNet(object):
self.cost = tf.reduce_mean(xent, name='xent') self.cost = tf.reduce_mean(xent, name='xent')
self.cost += self._decay() self.cost += self._decay()
moving_avg = tf.train.ExponentialMovingAverage( tf.scalar_summary('cost', self.cost)
0.99, num_updates=self.global_step, name='moving_avg')
self._extra_train_ops.append(moving_avg.apply([self.cost]))
tf.scalar_summary('cost', moving_avg.average(self.cost))
def _build_train_op(self): def _build_train_op(self):
"""Build training specific ops for the graph.""" """Build training specific ops for the graph."""
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment