label=i)
plt.legend()
plt.grid(axis='both')
plt.show()
if config.btl_size == 2:
min_range, max_range = -2., 2.
n = 20
step = (max_range - min_range) / float(n)
with torch.no_grad():
lines = []
for v1 in np.arange(min_range, max_range, step):
# |z| = (20, 2)
z = torch.stack([
torch.FloatTensor([v1] * n),
torch.FloatTensor([v2 for v2 in np.arange(min_range, max_range, step)]),
], dim=-1)
line = torch.clamp(model.decoder(z).view(n, 28, 28), 0, 1) # decoder(|Z|) = (20, 784) -> (20, 28, 28)
line = torch.cat([line[i] for i in range(n - 1, 0, -1)], dim=0) # Hidden Space 에 표시된 내용과 같도록 할건데
# 이미지의 경우 좌상단이 가장 작은 값으로 표시됨
# -> 역순으로 for 문을 수행하는 이유임
lines += [line]
lines = torch.cat(lines, dim=-1)
plt.figure(figsize=(20, 20))
show_image(lines)
if args.cuda:
autoencoder = autoencoder.cuda()
for epoch in range(1, args.epochs + 1):
autoencoder.train()
train_loss = 0
mnist_data = list(iter(train_loader))
for batch_idx in range(0, 1000):
data = torch.FloatTensor(mnist_data[batch_idx][0])
data = Variable(data)
if args.cuda:
data = data.cuda()
optimizer.zero_grad()
recon_batch, mu, logvar, encoded_rep = autoencoder(data)
if args.hmc:
init_x = encoded_rep.data.cpu().numpy()
hmc.get_hmc_sample(init_x, data, gpu=args.cuda)
optimizer.step()
if epoch % args.evaluate_interval == 0:
print("evaluating model")
evaluate_autoencoder(epoch)
sample = Variable(torch.randn(1, args.output_size))
if args.cuda:
sample = sample.cuda()
sample = autoencoder.decoder(sample).cpu()
save_image(sample.data.view(-1, 1, args.input_size, args.input_size), \
'results/sample_'+str(epoch)+'.png')