sharonFogel

It's correct, we just haven't updated the dataset dictionary after changing the data generation code.

I just looked at the code, I think you're right and there should be `self.netG.zero_grad()` between the first and second backprop. The third one is performed without gradient accumulation just...