What is the difference between real data and oracle data?

[sherlock1987]

[xuehui0725]

I'm confused about that, too. Look forward to your reply

[bhashithe]

I think its because you cannot really create the full distribution of a language when we create the embedding or word index distribution will loose some of the perspective. To handle that, in SeqGAN paper, they introduced a generated distribution which we know the full distribution of to train GANs.

So the difference is that, in Oracle data we know the full distribution while in real data we may not.