Closed AlexHex7 closed 6 years ago
I think b/c when we train the discriminator we train it on two batches (one real, one generated). When we train the generator we only train it on one batch (one to trick the discriminator). So since we already update D twice for each generator update I multiplied by 2.
@mjdietzx Oh, I see. Thanks!
in https://github.com/wayaai/SimGAN/blob/master/sim-gan.py#L309 print('Discriminator model loss real: {}.'.format(disc_loss_real / (log_interval k_d 2))) I think the 'k_d' should not *2
In the paper, it set kd = 1, kg = 2.
But in the code:
k_g = 2 # number of generative network updates per step
for _ in range(k_g * 2):
I think the 'k_g' should not * 2.