def translate_z(zs, model):
if isinstance(zs, np.ndarray):
zs = torch.from_numpy(zs).float()
running_time = get_current_time()
save_path_prefix = 'results/translation_z/' + running_time + '/'
sample = model.decode(zs).cpu()
if not os.path.exists(save_path_prefix):
os.makedirs(save_path_prefix)
save_image(sample.view(zs.shape[0], 1, args.img_size, args.img_size),
save_path_prefix + 'z_translation' + '.png')
def __init__(self, G, FE, D, Q, to_inverse, is_debugging=False):
has_cuda = torch.cuda.is_available()
self.device = torch.device("cuda" if has_cuda else "cpu")
self.G = G.to(self.device)
self.FE = FE.to(self.device)
self.D = D.to(self.device)
self.Q = Q.to(self.device)
self.batch_size = 100
self.current_time = gm.get_current_time()
self.to_inverse = to_inverse
self.saving_path = './results/' + str(self.to_inverse) + '/' + self.current_time
self.is_debugging = is_debugging
if not os.path.exists(self.saving_path) and not self.is_debugging:
os.makedirs(self.saving_path)
self.transform_pad = 4
self.img_size = 28
bhx = Variable(torch.zeros(X_dim), requires_grad=True)
def P(z):
h = nn.relu(z @ Wzh + bzh.repeat(z.size(0), 1))
X = nn.sigmoid(h @ Whx + bhx.repeat(h.size(0), 1))
return X
# =============================== TRAINING ====================================
params = [Wxh, bxh, Whz_mu, bhz_mu, Whz_var, bhz_var, Wzh, bzh, Whx, bhx]
solver = optim.Adam(params, lr=lr)
running_time = get_current_time()
for it in range(100000):
X, _ = mnist.train.next_batch(mb_size, shuffle=True)
print('X: ', X.shape)
# plot_a_numpy_array(X[0])
X = Variable(torch.from_numpy(X))
# Forward
z_mu, z_var = Q(X)
z = sample_z(z_mu, z_var)
X_sample = P(z)
# Loss
recon_loss = nn.binary_cross_entropy(X_sample, X,
size_average=False) / mb_size