WGAN_value_fct {RGAN} | R Documentation |
Implements the Wasserstein GAN (WGAN) value function as a function to be called in gan_trainer. Note that for this to work properly you also need to implement a weight clipper (or other procedure) to constrain the Discriminator weights.
WGAN_value_fct(real_scores, fake_scores)
real_scores |
The discriminator scores on real examples ($D(x)$) |
fake_scores |
The discriminator scores on fake examples ($D(G(z))$) |
The function returns a named list with the entries d_loss and g_loss