• Created by Unknown User (davidsch), last modified on Jul 03, 2016

We just add evaluation of the training steps, note use of session.run to evaluate softmax of logits

code:

ex05_tf_train.py

Notes:

Model does not seem to train well. Issues without batch normalization:

  • learning rate
  • parameter initialization
  • 'swings', all samples classified as A or B
  • How much data are we training on?

 

 

 

  • No labels