Uses more and more RAM with each passing experiment
zomseffen opened this issue · comments
I noticed that the training progress was using more and more RAM with each passing experiment.
After a quick search I found that this could be avoided by adding tf.reset_default_graph() at the end of the try_net method in the training.py .
With this change the code should be able to run on machines with way less available memory.
Thanks! I don't know why I removed that line at some point!