idx = int(random.random() * x_test.size(0))
recon = model(x_test[idx].view(1, -1)).squeeze()
show_image(x_test[idx])
show_image(recon)
# Hidden Space
if config.btl_size == 2:
color_map = ['brown', 'red', 'orange', 'yellow', 'green',
'blue', 'navy', 'purple', 'gray','black', ]
plt.figure(figsize=(20, 10))
with torch.no_grad():
latents = model.encoder(x_test[:1000])
for i in range(10):
target_latents = latents[y_test[:1000] == i]
y_target = y_test[:1000][y_test[:1000] == i]
plt.scatter(target_latents[:, 0],
target_latents[:, 1],
marker='o',
color=color_map[i],
label=i)
plt.legend()
plt.grid(axis='both')
plt.show()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.grid()
plt.savefig('./{}/loss.png'.format(log_dir))
# visualize latent space
test_dataset = MNIST(data_dir,
download=True,
train=False,
transform=img_transform)
test_loader = DataLoader(test_dataset, batch_size=10000, shuffle=False)
x, labels = iter(test_loader).next()
x = x.view(x.size(0), -1)
if use_gpu:
x = Variable(x).cuda()
z = model.encoder(x).cpu().data.numpy()
else:
x = Variable(x)
z = model.encoder(x).data.numpy()
plt.figure(figsize=(10, 10))
plt.scatter(z[:, 0], z[:, 1], marker='.', c=labels.numpy(), cmap=plt.cm.jet)
plt.colorbar()
plt.grid()
plt.savefig('./{}/latent_space.png'.format(log_dir))
# save result
np.save('./{}/loss_list.npy'.format(log_dir), np.array(loss_list))
torch.save(model.state_dict(), './{}/model_weights.pth'.format(log_dir))