lam = 10.0/(reduce(lambda x,y:x*y, gan.output_shape()))
dist = (gan.generator.sample - self.gan.inputs.sample).abs().mean()
​
dl = - d_real + d_fake
d_norm = 10 * dist
d_loss = ( dl + 0.5 * dl**2 / d_norm).mean()
​
g_loss = d_real - d_fake

{                                                                                       
  "class": "function:hypergan.losses.qp_loss.QPLoss"
}