Add option to include global step (passed in train_params) to explicitly pass to optimizer class if need be (sometimes you may want this in your optimizers, whereby you reset a variable or update it based on the global step value), since you cannot do this automatically with tf.train.get_global_step().
Add option to include global step (passed in train_params) to explicitly pass to optimizer class if need be (sometimes you may want this in your optimizers, whereby you reset a variable or update it based on the global step value), since you cannot do this automatically with tf.train.get_global_step().