requires_grad=True)
f_mid = dis_net(samples_mid)
# [0]:outputs tuple
# torch.ones(): grad can be implicitly created only for scalar outputs
grad_mid = grad(f_mid,
samples_mid,
torch.ones(f_mid.size()).cuda(),
create_graph=True)[0]
grad_mid_norm = grad_mid.norm(dim=1)
diff = (grad_mid_norm - 1).pow(2)
f_true = dis_net(samples_true)
f_fake = dis_net(samples_fake)
loss_critic = f_fake.mean() - f_true.mean() + \
(config['lda']*diff).mean()
prog.add_loss_critic(loss_critic.data.cpu().numpy()[0])
dis_optim.zero_grad()
loss_critic.backward()
dis_optim.step()
prog.add_loss_dis()
z = sampler.sample_z(config['dim_z'],
batchsize,
gaussian=config['gaussian'])
z = Variable(torch.from_numpy(z).cuda())
samples_fake = gen_net(z)
samples_fake /= config['scale']
f_fake = dis_net(samples_fake)
loss_gen = -f_fake.mean()
prog.add_loss_gen(loss_gen.data.cpu().numpy()[0])
config['num_mixture'],
scale=config['scale'],
std=config['std'])
samples_true /= config['scale']
samples_true = Variable(torch.from_numpy(samples_true))
z = sampler.sample_z(config['dim_z'],
batchsize,
gaussian=config['gaussian'])
z = Variable(torch.from_numpy(z))
samples_fake = gen_net(z).detach()
samples_fake /= config['scale']
f_true = dis_net(samples_true)
f_fake = dis_net(samples_fake)
loss_critic = f_fake.mean() - f_true.mean()
prog.add_loss_critic(loss_critic)
dis_optim.zero_grad()
loss_critic.backward()
dis_optim.step()
dis_net.clip()
prog.add_loss_dis()
z = sampler.sample_z(config['dim_z'],
batchsize,
gaussian=config['gaussian'])
z = Variable(torch.from_numpy(z))
samples_fake = gen_net(z)
samples_fake /= config['scale']
f_fake = dis_net(samples_fake)
loss_gen = -f_fake.mean()
(samples_fake_numpy.T*eps2).T
samples_mid = Variable(cuda(torch.from_numpy(
samples_mid_numpy)),requires_grad=True)
f_mid = dis_net(samples_mid)
# [0]:outputs tuple
# torch.ones(): grad can be implicitly created only for scalar outputs
grad_mid = grad(f_mid,samples_mid,cuda(torch.ones(f_mid.size())),
create_graph=True)[0]
grad_mid_norm = grad_mid.norm(dim=1)
diff = (grad_mid_norm - 1).pow(2)
f_true = dis_net(samples_true)
f_fake = dis_net(samples_fake)
loss_critic = f_fake.mean() - f_true.mean() + \
(config['lda']*diff).mean()
prog.add_loss_critic(cpu(loss_critic.data).numpy()[0])
dis_optim.zero_grad()
loss_critic.backward()
dis_optim.step()
prog.add_loss_dis()
z = sampler.sample_z(batchsize,config['dim_z'])
z = Variable(cuda(torch.from_numpy(z)))
samples_fake = gen_net(z)
f_fake = dis_net(samples_fake)
loss_gen = -f_fake.mean()
prog.add_loss_gen(cpu(loss_gen.data).numpy()[0])
gen_optim.zero_grad()
loss_gen.backward()
