sparkflow
sparkflow copied to clipboard
TF1.15 compatibility proposal
I cloned sparkflow and run with my TF1.15. The runtime log multiple errors which relate to sparkflow So i make some changes toward sparkflow on master branch and all the issue is gone.
I also change from using tf.get_collection(tf.GraphKeys.LOSSES)[0]
to tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
due to the fact that my model is written in old traditional way using tf.nn
instead of tf.layers