# g_scheduler = torch.optim.lr_scheduler.StepLR(
# g_optim, step_size=1, gamma=0.98)
# d_scheduler = torch.optim.lr_scheduler.StepLR(
# d_optim, step_size=1, gamma=0.98)
epoch = 0
idx_iter = 0
lst_eta = [generator.get_numpy_eta()]
while True:
idx_iter += 1
if isinstance(args.kappa, dict):
if epoch in args.kappa.keys():
discriminator.kappa = args.kappa[epoch]
print("Set kappa to {}".format(discriminator.kappa))
del args.kappa[epoch]
# XXX: note that training does not stop exactly at the end of the epoch
lst_d_loss, lst_g_loss = train_one_round(
loss_obj,
discriminator,
generator,
d_optim,
g_optim,
data_loader_iter,
noise_generator,
fake_batch_size=args.fake_batch_size,
device=None,
