Performance issues in lib/model/train_val.py

[JamesCao2048]

Hello, I found that in the function train_model_random, lib/model/train_val.py, tf.assign will create redundant node in tf computation graph repeatedly(as it is in a while loop). I think tf.assign should be called only once before the loop to create a op, and use session to run this op repeatedly.

The same issue exists here.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.