Logits and log_probs set to 0 when using cross_entropy

[datduonguva]

In MaskGan, seq2seq_vd, if the training strategy is cross_entropy, all the logits, ,log_probs output of the generator's decoder are set to 0. Why is this the case?

models/research/maskgan/models/seq2seq_vd.py

Line 354 in 0672bdc

if FLAGS.gen_training_strategy != 'cross_entropy':

@a-dai