In DCGAN example, when training G, D is not freezed

[fxia22]

Unlike in GAN paper, there is no

for p in netD.parameters():
    p.requires_grad = False # to avoid computation

when updating generator. Is this on purpose or by mistake?

In @soumith's torch reference implementation, D is fixed when updating G:

   local df_do = criterion:backward(output, label)
   local df_dg = netD:updateGradInput(input, df_do)
   netG:backward(noise, df_dg)

[soumith]

oh, i kinda didn't do it in hindsight, I should do it.
However that will only save computation.
D is still frozen because I dont call optimizerD.step() (which is when D's parameters actually get updated)

[fxia22]

Thanks, that makes sense.