Problems with variance_scaling_initializer
Closed this issue · 3 comments
I have run into two issues with line 59.
When run as shown, I get:
AttributeError: module 'tensorflow' has no attribute 'variance_scaling_initializer'
If I change the line to read
weight_initializer = tf.contrib.layers.variance_scaling_initializer(mode="fan_avg", distribution="uniform", scale=sigma)
I then get this error:
TypeError: variance_scaling_initializer() got an unexpected keyword argument 'distribution'
I only installed TensorFlow last week, and your file is only 23 days old, so I doubt that we have a version issue. Do these errors mean anything to you?
If I change the line to (note FAN_AVG is now in caps):
weight_initializer = tf.contrib.layers.variance_scaling_initializer(mode="FAN_AVG")
it will run, but the epoch graphs are empty.
I just run it with tensorflow 1.4.1 (from pip --upgrade) and the original version seems to be working fine
Code is properly working with current TensorFlow version (1.4.1).