Open chrismessenger opened 4 years ago
The convolutional weight and bias names have historical + '1' added to them which is not required. This is also the case in the q and r2 networks.
Plus, there are lots of commented out code that can get deleted when creating the weights.
Also, the
self.nonlinearity_mean = tf.clip_by_value
is never used so can be removed.
You can also delete the commented out "make positive definite covariance matrix code"
The convolutional weight and bias names have historical + '1' added to them which is not required. This is also the case in the q and r2 networks.
Plus, there are lots of commented out code that can get deleted when creating the weights.