Question about Rollout

[lantaoyu]

Question about Rollout

[nathan-whitaker]

I think I found what I was looking for.

SeqGAN/rollout.py

Line 58 in 5f2c0a5

next_token = tf.cast(tf.reshape(tf.multinomial(log_prob, 1), [self.batch_size]), tf.int32)

This line contains a call to https://www.tensorflow.org/api_docs/python/tf/multinomial

Which performs a sample over the logits generated from the network instead of taking the max like the generator network does.

[guotong1988]

Great investigation！

[guotong1988]

I still don't know exactly what N Time Monte Carlo sampling is.. Could you please explain? Thank you @LantaoYu @nathan-whitaker