def visualize_progession(self, i, iters, epoch, errD, errG, D_x, D_G_z1,
D_G_z2):
"""
Handle the printing and plotting of the loss
:return:
"""
# Output training stats
if i:
print(
'[%d/%d][%d/%d]\tLoss_D: %.4f\tLoss_G: %.4f\tD(x): %.4f\tD(G(z)): %.4f / %.4f'
% (epoch, num_epochs, i, len(self.dataloader), errD.item(),
errG.item(), D_x, D_G_z1, D_G_z2))
# Visualization with visdom
viz.line([errD.item()], [iters],
win='ErrorD',
update='append' if iters > 0 else None,
name='Loss Discriminator',
opts=dict(xlabel='Step', ylabel='Loss', title='Losses'))
# Visualization with visdom
viz.line([errG.item()], [epoch],
win='ErrorD',
update='append' if iters > 0 else None,
name='Loss Generator',
opts=dict(xlabel='Step', ylabel='Loss', title='Losses'))
# Check how the generator is doing by saving G's output on fixed_noise
if (iters % 20 == 0) or ((epoch == num_epochs - 1) and
(i == len(self.dataloader) - 1)):
with torch.no_grad():
fake = self.netG(self.fixed_noise).detach().cpu()
display_city(fake[0], win_name='Test Example')
def save_observations(self, epoch):
"""
Save the n_observations generated cities for a given layer
:param n_observations: number of observations to save
"""
logging.info('Save {} observations at epoch {}'.format(
self.n_observations, epoch))
with torch.no_grad():
fake = self.netG(self.fixed_noise).detach().cpu()
for i in range(self.n_observations):
display_city(fake[i],
win_name='Test Example',
folder_name=self.experiment_name,
epoch='observation_{}_epoch_{}'.format(i, epoch))
def visualize_progession(self, epoch, loss_discriminator, loss_generator,
G1, D):
"""
Handle the printing and plotting of the loss
:return:
"""
# Output training stats
logging.info(
'[{}/{}]\tLoss_D: {:.4f}\tLoss_G: {:.4f} %Fake as Real: {:.1f} %Real as real: {:.1f}'
.format(epoch, self.num_epochs + self.start_epoch,
loss_discriminator, loss_generator, G1 * 100, D * 100))
if self.viz:
# Visualization with visdom
self.viz.line([loss_discriminator], [epoch],
win='ErrorD',
update='append' if epoch > 0 else None,
name='Loss Discriminator',
opts=dict(xlabel='Step',
ylabel='Loss',
title='Losses'))
# Visualization with visdom
self.viz.line([loss_generator], [epoch],
win='ErrorD',
update='append' if epoch > 0 else None,
name='Loss Generator',
opts=dict(xlabel='Step',
ylabel='Loss',
title='Losses'))
# Check how the generator is doing by saving G's output on fixed_noise
if (epoch % 10 == 0):
with torch.no_grad():
fake = self.netG(self.fixed_noise).detach().cpu()
display_city(fake[0],
win_name='Test Example',
viz=self.viz,
folder_name=self.experiment_name,
epoch=epoch)