I am programming GANs with Tensorflow. Now I am at the point I have to decide if I go through the data in epochs or in steps.

As I found out the epoch way is more efficient (Why do neural network researchers care about epochs?) I am still wondering how this works.

Suppose I want to train the discriminator 3 times (so 3 times updating the weights) and then the generator 1 time.

Do I need for each step a new minibatch?

How do you generally train GANs?


  • $\begingroup$ Recently I have uploaded collection of various GAN models on github repo. It is torch7 based, and very easy to run. The code is simple enough to understand with experimental results. Hope this will help github.com/nashory/gans-collection.torch $\endgroup$
    – nashory
    Sep 20, 2017 at 16:06

1 Answer 1


GAN training is still very much a black-art, so it's hard to give firm advice. In terms of using minibatches, there is a discussion of it in Section 3.2 in this paper.

I highly recommend watching the NIPS tutorial by Ian if you haven't already.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.