decoder = Decoder()
encoder.to(device)
decoder.to(device)
print("number of parameters: {}".format(encoder._n_parameters() +
decoder._n_parameters()))
train_hdf5_file = os.getcwd() + \
f'/Gaussian/data/training_set_64_gaussian1_25000.hdf5'
train_loader = load_data_1scale(train_hdf5_file,
args.n_train,
args.batch_size,
singlescale=True)
optimizer = torch.optim.Adam(itertools.chain(encoder.parameters(),
decoder.parameters()),
lr=args.lr,
betas=(args.beta1, args.beta2))
def loss_function(recon_x, x, mu, logvar):
Recon_loss = F.mse_loss(recon_x.view(-1, 4096),
x.view(-1, 4096),
size_average=False)
mu = mu.reshape(-1, 256)
logvar = logvar.reshape(-1, 256)
KLD = torch.sum(-0.5 *
torch.sum(1 + logvar - mu.pow(2) - logvar.exp(), dim=1),
dim=0)
return Recon_loss + args.beta_vae * KLD, Recon_loss, KLD
encoder.to(device)
decoder.to(device)
print("number of parameters: {}".format(encoder._n_parameters()+decoder._n_parameters()))
train_hdf5_file = os.getcwd() + \
f'/Gaussian/data/training_set_16_gaussian.hdf5'
test_hdf5_file = os.getcwd() + \
f'/Gaussian/data/test_set_16_gaussian.hdf5'
train_loader = load_data_1scale(train_hdf5_file, args.n_train, args.batch_size,singlescale=True)
with h5py.File(test_hdf5_file, 'r') as f:
x_test = f['test'][()]
x_test =x_test
optimizer= torch.optim.Adam(
itertools.chain(encoder.parameters(), decoder.parameters()), lr=args.lr, betas=(args.beta1, args.beta2))
def test(epoch,x_test):
encoder.eval()
decoder.eval()
z = torch.randn(9, 1, 4 ,4).to(device)
imgs = decoder(z)
samples = np.squeeze(imgs.data.cpu().numpy())
plot_generation(samples,epoch,output_dir,1)
real_imgs = x_test[[10,30,50,100]]
real_imgs = (torch.FloatTensor(real_imgs)).to(device)
encoded_imgs,_,_ = encoder(real_imgs)
decoded_imgs = decoder(encoded_imgs)
samples_gen = np.squeeze(decoded_imgs.data.cpu().numpy())
samples_real = np.squeeze(real_imgs.data.cpu().numpy())
