def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['E_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
if torch.cuda.is_available():
self.y_real_, self.y_fake_ = Variable(
torch.ones(self.batch_size, 1).cuda()), Variable(
torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(
torch.ones(self.batch_size,
1)), Variable(torch.zeros(self.batch_size, 1))
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
# reset training mode of G and E
self.G.train()
self.E.train()
epoch_start_time = time.time()
for iter, (X, _) in enumerate(self.data_loader):
X = utils.to_var(X)
"""Discriminator"""
z = utils.to_var(
torch.randn((self.batch_size,
self.z_dim)).view(-1, self.z_dim, 1, 1))
X_hat = self.G(z)
D_real = self.D(X).squeeze().view(-1, 1)
D_fake = self.D(X_hat).squeeze().view(-1, 1)
D_loss = self.BCE_loss(D_real, self.y_real_) + self.BCE_loss(
D_fake, self.y_fake_)
self.train_hist['D_loss'].append(D_loss.data[0])
# Optimize
D_loss.backward()
self.D_optimizer.step()
self.__reset_grad()
"""Encoder"""
z = utils.to_var(
torch.randn((self.batch_size,
self.z_dim)).view(-1, self.z_dim, 1, 1))
X_hat = self.G(z)
z_mu, z_sigma = self.E(X_hat)
z_mu, z_sigma = z_mu.squeeze(), z_sigma.squeeze()
# - loglikehood
E_loss = torch.mean(
torch.mean(
0.5 * (z - z_mu)**2 * torch.exp(-z_sigma) +
0.5 * z_sigma + 0.5 * np.log(2 * np.pi), 1))
self.train_hist['E_loss'].append(E_loss.data[0])
# Optimize
E_loss.backward()
self.E_optimizer.step()
self.__reset_grad()
"""Generator"""
# Use both Discriminator and Encoder to update Generator
z = utils.to_var(
torch.randn((self.batch_size,
self.z_dim)).view(-1, self.z_dim, 1, 1))
X_hat = self.G(z)
D_fake = self.D(X_hat).squeeze().view(-1, 1)
z_mu, z_sigma = self.E(X_hat)
z_mu, z_sigma = z_mu.squeeze(), z_sigma.squeeze()
mode_loss = torch.mean(
torch.mean(
0.5 * (z - z_mu)**2 * torch.exp(-z_sigma) +
0.5 * z_sigma + 0.5 * np.log(2 * np.pi), 1))
G_loss = self.BCE_loss(D_fake, self.y_real_)
total_loss = G_loss + mode_loss
self.train_hist['G_loss'].append(G_loss.data[0])
# Optimize
total_loss.backward()
self.G_optimizer.step()
self.__reset_grad()
""" Plot """
if (iter + 1
) == self.data_loader.dataset.__len__() // self.batch_size:
# Print and plot every epoch
print(
'Epoch-{}; D_loss: {:.4}; G_loss: {:.4}; E_loss: {:.4}\n'
.format(epoch, D_loss.data[0], G_loss.data[0],
E_loss.data[0]))
for iter, (X, _) in enumerate(self.valid_loader):
X = utils.to_var(X)
self.visualize_results(X, epoch + 1)
break
break
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
# Save model every 5 epochs
if epoch % 5 == 0:
self.save()
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save final training results")
self.save()
# Generate animation of reconstructed plot
utils.generate_animation(
self.root + '/' + self.result_dir + '/' + self.dataset + '/' +
self.model_name + '/reconstructed', self.epoch)
utils.loss_plot(
self.train_hist,
os.path.join(self.root, self.save_dir, self.dataset,
self.model_name), self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
if self.gpu_mode:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1).cuda()), Variable(torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1)), Variable(torch.zeros(self.batch_size, 1))
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__() // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = Variable(x_.cuda()), Variable(z_.cuda())
else:
x_, z_ = Variable(x_), Variable(z_)
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = self.BCE_loss(D_real, self.y_real_)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = self.BCE_loss(D_fake, self.y_fake_)
D_loss = D_real_loss + D_fake_loss
self.train_hist['D_loss'].append(D_loss.data[0])
D_loss.backward()
self.D_optimizer.step()
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = self.BCE_loss(D_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1), (iter + 1), self.data_loader.dataset.__len__() // self.batch_size, D_loss.data[0], G_loss.data[0]))
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.visualize_results((epoch+1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
self.y_real_, self.y_fake_ = torch.ones(self.batch_size,
1), torch.zeros(
self.batch_size, 1)
if self.gpu_mode:
self.y_real_, self.y_fake_ = self.y_real_.cuda(
), self.y_fake_.cuda()
self.D.train()
print('training start!!')
start_time = time.time()
Q_stack = []
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__(
) // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = x_.cuda(), z_.cuda()
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = self.BCE_loss(D_real, self.y_real_)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = self.BCE_loss(D_fake, self.y_fake_)
D_loss = D_real_loss + D_fake_loss
self.train_hist['D_loss'].append(D_loss.item())
D_loss.backward()
self.D_optimizer.step()
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
######## following codes are practical implementation for the paper ##########
pertubation_del = (torch.randn(self.batch_size,
self.z_dim)).cuda()
eps = self.eps
pertu_length = torch.norm(pertubation_del, dim=1, keepdim=True)
pertubation_del = (pertubation_del / pertu_length) * eps
z_prime = z_ + pertubation_del
pertube_images = self.G(z_) - self.G(z_prime)
pertube_latent_var = z_ - z_prime
Q = torch.norm(pertube_images.view(
self.batch_size, -1), dim=1) / torch.norm(
pertube_latent_var.view(self.batch_size, -1), dim=1)
print(Q)
L_max = 0.0
L_min = 0.0
count_max = 0
count_min = 0
for i in range(self.batch_size):
if Q[i] > self.eig_max:
L_max += (Q[i] - self.eig_max)**2
count_max += 1
if Q[i] < self.eig_min:
L_min += (Q[i] - self.eig_min)**2
count_min += 1
L = L_max + L_min
#################### end of implementation for the paper ####################
G_loss = self.BCE_loss(D_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.item())
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1),
(iter + 1), self.data_loader.dataset.__len__() //
self.batch_size, D_loss.item(), G_loss.item()))
print(L)
print(count_max)
print(count_min)
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
with torch.no_grad():
self.gen_mode(4, epoch)
self.visualize_results((epoch + 1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(
self.result_dir + '/' + self.dataset + '/' + self.model_name +
'/' + self.model_name, self.epoch)
utils.loss_plot(
self.train_hist,
os.path.join(self.save_dir, self.dataset, self.model_name),
self.model_name)
def train(self):
self.G.apply(self.G.weights_init)
print(' training start!! (no conditional)')
start_time = time.time()
for classe in range(10):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
# self.G.apply(self.G.weights_init) does not work for instance
del self.E
self.E = Encoder(self.z_dim, self.dataset, self.conditional)
self.E_optimizer = optim.Adam(
self.E.parameters(),
lr=self.lr) #, lr=args.lrD, betas=(args.beta1, args.beta2))
if self.gpu_mode:
self.E.cuda(self.device)
best = 100000
self.data_loader_train = get_iter_dataset(self.dataset_train,
self.list_class_train,
self.batch_size, classe)
self.data_loader_valid = get_iter_dataset(self.dataset_valid,
self.list_class_valid,
self.batch_size, classe)
early_stop = 0.
for epoch in range(self.epoch):
epoch_start_time = time.time()
# print("number of batch data")
# print(len(self.data_loader_train))
self.E.train()
self.G.train()
sum_loss_train = 0.
n_batch = 0.
#for iter in range(self.size_epoch):
for iter, (x_, t_) in enumerate(self.data_loader_train):
n_batch += 1
#x_ = sort_utils.get_batch(list_classes, classe, self.batch_size)
#x_ = torch.FloatTensor(x_)
x_ = Variable(x_)
if self.gpu_mode:
x_ = x_.cuda(self.device)
# VAE
z_, mu, logvar = self.E(x_)
recon_batch = self.G(z_)
# train
self.G_optimizer.zero_grad()
self.E_optimizer.zero_grad()
g_loss = self.loss_function(recon_batch, x_, mu, logvar)
g_loss.backward() #retain_variables=True)
sum_loss_train += g_loss.data[0]
self.G_optimizer.step()
self.E_optimizer.step()
self.train_hist['D_loss'].append(g_loss.data[0])
self.train_hist['G_loss'].append(g_loss.data[0])
if ((iter + 1) % 100) == 0:
print(
"classe : [%1d] Epoch: [%2d] [%4d/%4d] G_loss: %.8f, E_loss: %.8f"
% (classe, (epoch + 1),
(iter + 1), self.size_epoch, g_loss.data[0],
g_loss.data[0]))
sum_loss_train = sum_loss_train / np.float(n_batch)
sum_loss_valid = 0.
n_batch = 0.
n_batch = 1.
self.E.eval()
self.G.eval()
for iter, (x_, t_) in enumerate(self.data_loader_valid):
n_batch += 1
max_val, max_indice = torch.max(t_, 0)
mask_idx = torch.nonzero(t_ == classe)
if mask_idx.dim() == 0:
continue
x_ = torch.index_select(x_, 0, mask_idx[:, 0])
t_ = torch.index_select(t_, 0, mask_idx[:, 0])
if self.gpu_mode:
x_ = Variable(x_.cuda(self.device), volatile=True)
else:
x_ = Variable(x_)
# VAE
z_, mu, logvar = self.E(x_)
recon_batch = self.G(z_)
G_loss = self.loss_function(recon_batch, x_, mu, logvar)
sum_loss_valid += G_loss.data[0]
sum_loss_valid = sum_loss_valid / np.float(n_batch)
print(
"classe : [%1d] Epoch: [%2d] Train_loss: %.8f, Valid_loss: %.8f"
% (classe, (epoch + 1), sum_loss_train, sum_loss_valid))
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
self.visualize_results((epoch + 1), classe)
if sum_loss_valid < best:
best = sum_loss_valid
self.save_G(classe)
early_stop = 0.
# We dit early stopping of the valid performance doesn't
# improve anymore after 50 epochs
if early_stop == 150:
break
else:
early_stop += 1
result_dir = self.result_dir + '/' + 'classe-' + str(classe)
utils.generate_animation(result_dir + '/' + self.model_name,
epoch + 1)
utils.loss_plot(self.train_hist, result_dir, self.model_name)
np.savetxt(
os.path.join(result_dir,
'vae_training_' + self.dataset + '.txt'),
np.transpose([self.train_hist['G_loss']]))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
self.y_real_, self.y_fake_ = torch.ones(self.batch_size,
1), torch.zeros(
self.batch_size, 1)
if self.gpu_mode:
self.y_real_, self.y_fake_ = self.y_real_.cuda(
), self.y_fake_.cuda()
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__(
) // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = x_.cuda(), z_.cuda()
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = -torch.mean(D_real)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = torch.mean(D_fake)
# gradient penalty
alpha = torch.rand((self.batch_size, 1, 1, 1))
if self.gpu_mode:
alpha = alpha.cuda()
x_hat = alpha * x_.data + (1 - alpha) * G_.data
x_hat.requires_grad = True
pred_hat = self.D(x_hat)
if self.gpu_mode:
gradients = grad(outputs=pred_hat,
inputs=x_hat,
grad_outputs=torch.ones(
pred_hat.size()).cuda(),
create_graph=True,
retain_graph=True,
only_inputs=True)[0]
else:
gradients = grad(outputs=pred_hat,
inputs=x_hat,
grad_outputs=torch.ones(pred_hat.size()),
create_graph=True,
retain_graph=True,
only_inputs=True)[0]
gradient_penalty = self.lambda_ * (
(gradients.view(gradients.size()[0], -1).norm(2, 1) - 1)**
2).mean()
D_loss = D_real_loss + D_fake_loss + gradient_penalty
D_loss.backward()
self.D_optimizer.step()
if ((iter + 1) % self.n_critic) == 0:
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = -torch.mean(D_fake)
self.train_hist['G_loss'].append(G_loss.item())
G_loss.backward()
self.G_optimizer.step()
self.train_hist['D_loss'].append(D_loss.item())
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1),
(iter + 1), self.data_loader.dataset.__len__() //
self.batch_size, D_loss.item(), G_loss.item()))
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
with torch.no_grad():
self.visualize_results((epoch + 1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
# utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + '%03d'% self.epoch +'/' + self.model_name + '/'
# + self.model_name,self.epoch)
utils.loss_plot(
self.train_hist,
os.path.join(self.save_dir, self.dataset, self.model_name),
self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
self.D.train()
self.E.train()
print('training start!!')
start_time = time.time()
first_time=True
self.accuracy_hist=[]
for epoch in range(self.epoch):
self.G.train() # check here!!
epoch_start_time = time.time()
decay=0.98**epoch
self.E_optimizer = optim.Adam(self.E.parameters(), lr=decay * 0.3 * self.args.lrD,
betas=(self.args.beta1, self.args.beta2))
self.G_optimizer = optim.Adam(self.G.parameters(), lr=decay * 3 * self.args.lrG, betas=(self.args.beta1, self.args.beta2))
self.D_optimizer = optim.Adam(self.D.parameters(), lr=decay * self.args.lrD, betas=(self.args.beta1, self.args.beta2))
for M_epoch in range(5):
for iter, (batch_x, batch_y) in enumerate(self.train_loader):
x_=batch_x
z_=torch.rand((self.batch_size, self.z_dim))
x_, z_ = Variable(x_.cuda()), Variable(z_.cuda())
G_batch_size = batch_x.size()[0]
if G_batch_size < self.batch_size:
break
# x_ (batch, 1L, 28L, 28L)
# z_ (batch, 62L)
# update D network:
image_real = Variable(batch_x.cuda())
self.E.eval()
y_real = self.E(image_real)
y_real = nn.Softmax()(y_real)
y_real = (y_real).data.cpu().numpy() #
self.D_optimizer.zero_grad()
D_real = self.D(x_)
if first_time:
y_real = (1 / float(self.class_num)) * np.ones((G_batch_size, self.class_num)) # first_time
y_real = np.concatenate((y_real, 2*np.ones((np.shape(y_real)[0], 1))), axis=1)
ones=np.ones((np.shape(y_real)[0],np.shape(y_real)[1]))
ones[:,-1]=0
ones=torch.FloatTensor(ones)
ones=Variable(ones).cuda()
y_real=torch.FloatTensor(y_real).cuda()
D_real_loss = torch.nn.BCEWithLogitsLoss(weight=y_real)(D_real,ones)
G_input, conditional_label = self.gen_cond_label(self.batch_size)
G_ = self.G(G_input, 0)
D_fake = self.D(G_)
y_fake_1 = np.tile(np.zeros((self.class_num)), (self.batch_size, 1))
y_fake_2 = np.tile(np.ones((1)), (self.batch_size, 1))
y_fake = np.concatenate((y_fake_1,y_fake_2),axis=1)
y_fake = Variable(torch.FloatTensor(y_fake).cuda())
D_fake_loss=torch.nn.BCEWithLogitsLoss()(D_fake,y_fake)
D_loss = D_real_loss + D_fake_loss
self.train_hist['D_loss'].append(D_loss.data[0])
D_loss.backward()
self.D_optimizer.step()
# update G network:
self.G_optimizer.zero_grad()
G_input, conditional_label = self.gen_cond_label(self.batch_size)
G_ = self.G(G_input, 0)
D_fake = self.D(G_)
G_y_real=np.concatenate((conditional_label.numpy(),np.tile([0],(self.batch_size,1))),axis=1)
G_y_real=Variable(torch.FloatTensor(G_y_real)).cuda()
G_loss=torch.nn.BCEWithLogitsLoss()(D_fake,G_y_real)
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1), (iter + 1), len(self.data_X) // self.batch_size, D_loss.data[0], G_loss.data[0]))
self.E_training(200)
first_time = False
self.visualize_results((epoch+1))
self.compute_accuracy()
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.save()
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
self.train_hist['D_norm'] = []
f = open("%s/results.txt" % self.log_dir, "w")
f.write("d_loss,g_loss,d_norm\n")
if self.gpu_mode:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1).cuda()), Variable(torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1)), Variable(torch.zeros(self.batch_size, 1))
#for iter, ((x1_,_), (x2_,_)) in enumerate(zip(self.data_loader, self.data_loader)):
# import pdb
# pdb.set_trace()
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__() // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = Variable(x_.cuda(), requires_grad=True), \
Variable(z_.cuda())
else:
x_, z_ = Variable(x_, requires_grad=True), \
Variable(z_)
# update D network
D_real = self.D(x_)
# compute gradient penalty
grad_wrt_x = grad(outputs=D_real, inputs=x_,
grad_outputs=torch.ones(D_real.size()).cuda(),
create_graph=True, retain_graph=True, only_inputs=True)[0]
g_norm = ((grad_wrt_x.view(grad_wrt_x.size()[0], -1).norm(2, 1) - 1) ** 2).mean()
self.train_hist['D_norm'].append(g_norm.data.item())
self.D_optimizer.zero_grad()
G_ = self.G(z_).detach()
alpha = float(np.random.random())
Xz = Variable(alpha*x_.data + (1.-alpha)*G_.data)
D_Xz = self.D(Xz)
D_loss = self.BCE_loss(D_Xz, alpha*self.y_real_)
self.train_hist['D_loss'].append(D_loss.data.item())
D_loss.backward()
self.D_optimizer.step()
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = self.BCE_loss(D_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.data.item())
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f, D_norm: %.8f" %
((epoch + 1),
(iter + 1),
self.data_loader.dataset.__len__() // self.batch_size,
D_loss.data.item(),
G_loss.data.item(),
g_norm.data.item()))
f.write("%.8f,%.8f,%.8f\n" % (D_loss.data.item(), G_loss.data.item(), g_norm.data.item()))
f.flush()
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.visualize_results((epoch+1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
f.close()
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
if self.gpu_mode:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1).cuda()), Variable(torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1)), Variable(torch.zeros(self.batch_size, 1))
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__() // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = Variable(x_.cuda()), Variable(z_.cuda())
else:
x_, z_ = Variable(x_), Variable(z_)
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = -torch.mean(D_real)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = torch.mean(D_fake)
D_loss = D_real_loss + D_fake_loss
D_loss.backward()
self.D_optimizer.step()
# clipping D
for p in self.D.parameters():
p.data.clamp_(-self.c, self.c)
if ((iter+1) % self.n_critic) == 0:
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = -torch.mean(D_fake)
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
self.train_hist['D_loss'].append(D_loss.data[0])
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1), (iter + 1), self.data_loader.dataset.__len__() // self.batch_size, D_loss.data[0], G_loss.data[0]))
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.visualize_results((epoch+1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)
def train(self):
train_hist_keys = [
'D2d_loss', 'D3d_loss', 'D2d_acc', 'D3d_acc', 'G_loss',
'per_epoch_time', 'total_time'
]
if 'recon' in self.loss_option:
train_hist_keys.append('G_loss_recon')
if 'dist' in self.loss_option:
train_hist_keys.append('G_loss_dist')
if not hasattr(self, 'epoch_start'):
self.epoch_start = 0
if not hasattr(self, 'train_hist'):
self.train_hist = {}
for key in train_hist_keys:
self.train_hist[key] = []
else:
existing_keys = self.train_hist.keys()
num_hist = [len(self.train_hist[key]) for key in existing_keys]
num_hist = max(num_hist)
for key in train_hist_keys:
if key not in existing_keys:
self.train_hist[key] = [0] * num_hist
print('new key added: {}'.format(key))
if self.gpu_mode:
self.y_real_ = Variable((torch.ones(self.batch_size, 1)).cuda())
self.y_fake_ = Variable((torch.zeros(self.batch_size, 1)).cuda())
else:
self.y_real_ = Variable((torch.ones(self.batch_size, 1)))
self.y_fake_ = Variable((torch.zeros(self.batch_size, 1)))
nPairs = self.batch_size * (self.batch_size - 1)
normalizerA = self.data_loader.dataset.muA / self.data_loader.dataset.stddevA # normalization
normalizerB = self.data_loader.dataset.muB / self.data_loader.dataset.stddevB # normalization
eps = 1e-16
self.D2d.train()
self.D3d.train()
start_time = time.time()
nBatchesPerEpoch = self.data_loader.dataset.__len__(
) // self.batch_size
print('training start from epoch {}!!'.format(self.epoch_start + 1))
for epoch in range(self.epoch_start, self.epoch):
self.G.train()
epoch_start_time = time.time()
start_time_epoch = time.time()
for iB, (x3D_, y_, x2D_) in enumerate(self.data_loader):
if iB == nBatchesPerEpoch:
break
projected, _ = torch.max(x3D_[:, 1:, :, :, :],
4,
keepdim=False)
x3D_ = x3D_[:, 0:1, :, :, :]
y_random_pcode_ = torch.floor(
torch.rand(self.batch_size) * self.num_c_expr).long()
y_random_pcode_onehot_ = torch.zeros(self.batch_size,
self.num_c_expr)
y_random_pcode_onehot_.scatter_(1, y_random_pcode_.view(-1, 1),
1)
y_id_ = y_['id']
y_pcode_ = y_['pcode']
y_pcode_onehot_ = torch.zeros(self.batch_size, self.num_c_expr)
y_pcode_onehot_.scatter_(1, y_pcode_.view(-1, 1), 1)
if self.gpu_mode:
x2D_ = Variable(x2D_.cuda())
x3D_ = Variable(x3D_.cuda())
projected = Variable(projected.cuda())
y_id_ = Variable(y_id_.cuda())
y_pcode_ = Variable(y_pcode_.cuda())
y_pcode_onehot_ = Variable(y_pcode_onehot_.cuda())
y_random_pcode_ = Variable(y_random_pcode_.cuda())
y_random_pcode_onehot_ = Variable(
y_random_pcode_onehot_.cuda())
else:
x2D_ = Variable(x2D_)
x3D_ = Variable(x3D_)
projected = Variable(projected)
y_id_ = Variable(y_id_)
y_pcode_ = Variable(y_pcode_)
y_pcode_onehot_ = Variable(y_pcode_onehot_)
y_random_pcode_ = Variable(y_random_pcode_)
y_random_pcode_onehot_ = Variable(y_random_pcode_onehot_)
# update D network
for iD in range(self.n_critic):
self.D2d_optimizer.zero_grad()
self.D3d_optimizer.zero_grad()
d_gan2d, d_id2d, d_expr2d = self.D2d(projected)
loss_d_real_gan2d = self.BCE_loss(d_gan2d, self.y_real_)
loss_d_real_id2d = self.CE_loss(d_id2d, y_id_)
loss_d_real_expr2d = self.CE_loss(d_expr2d, y_pcode_)
d_gan3d, d_id3d, d_expr3d = self.D3d(x3D_)
loss_d_real_gan3d = self.BCE_loss(d_gan3d, self.y_real_)
loss_d_real_id3d = self.CE_loss(d_id3d, y_id_)
loss_d_real_expr3d = self.CE_loss(d_expr3d, y_pcode_)
xhat2d, xhat3d = self.G(x2D_, y_random_pcode_onehot_)
d_fake_gan2d, _, _ = self.D2d(xhat2d)
d_fake_gan3d, _, _ = self.D3d(xhat3d)
loss_d_fake_gan2d = self.BCE_loss(d_fake_gan2d,
self.y_fake_)
loss_d_fake_gan3d = self.BCE_loss(d_fake_gan3d,
self.y_fake_)
num_correct_real2d = torch.sum(d_gan2d > 0.5)
num_correct_fake2d = torch.sum(d_fake_gan2d < 0.5)
D2d_acc = float(num_correct_real2d.data[0] +
num_correct_fake2d.data[0]) / (
self.batch_size * 2)
num_correct_real3d = torch.sum(d_gan3d > 0.5)
num_correct_fake3d = torch.sum(d_fake_gan3d < 0.5)
D3d_acc = float(num_correct_real3d.data[0] +
num_correct_fake3d.data[0]) / (
self.batch_size * 2)
D2d_loss = loss_d_real_gan2d + loss_d_real_id2d + loss_d_real_expr2d + loss_d_fake_gan2d
D3d_loss = loss_d_real_gan3d + loss_d_real_id3d + loss_d_real_expr3d + loss_d_fake_gan3d
if iD == 0:
self.train_hist['D2d_loss'].append(D2d_loss.data[0])
self.train_hist['D3d_loss'].append(D3d_loss.data[0])
self.train_hist['D2d_acc'].append(D2d_acc)
self.train_hist['D3d_acc'].append(D3d_acc)
D2d_loss.backward(retain_graph=True)
D3d_loss.backward()
if D2d_acc < 0.8:
self.D2d_optimizer.step()
if D3d_acc < 0.8:
self.D3d_optimizer.step()
# update G network
for iG in range(self.n_gen):
self.G_optimizer.zero_grad()
xhat2d, xhat3d = self.G(x2D_, y_pcode_onehot_)
d_gan2d, d_id2d, d_expr2d = self.D2d(xhat2d)
loss_g_gan2d = self.BCE_loss(d_gan2d, self.y_real_)
loss_g_id2d = self.CE_loss(d_id2d, y_id_)
loss_g_expr2d = self.CE_loss(d_expr2d, y_pcode_)
d_gan3d, d_id3d, d_expr3d = self.D3d(xhat3d)
loss_g_gan3d = self.BCE_loss(d_gan3d, self.y_real_)
loss_g_id3d = self.CE_loss(d_id3d, y_id_)
loss_g_expr3d = self.CE_loss(d_expr3d, y_pcode_)
G_loss = loss_g_gan2d + loss_g_id2d + loss_g_expr2d + \
loss_g_gan3d + loss_g_id3d + loss_g_expr3d
if iG == 0:
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
if ((iB + 1) % 10) == 0 or (iB + 1) == nBatchesPerEpoch:
secs = time.time() - start_time_epoch
hours = secs // 3600
mins = secs / 60 % 60
#print("%2dh%2dm E[%2d] B[%d/%d] D: %.4f, G: %.4f, D_acc:%.4f"%
print(
"%2dh%2dm E[%2d] B[%d/%d] D: %.4f/%.4f, G: %.4f, D_acc:%.4f/%.4f"
% (hours, mins, (epoch + 1),
(iB + 1), nBatchesPerEpoch, D2d_loss.data[0],
D3d_loss.data[0], G_loss.data[0], D3d_acc, D2d_acc))
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
if epoch == 0 or (epoch + 1) % 5 == 0:
self.dump_x_hat(xhat2d, xhat3d, epoch + 1)
self.save()
utils.loss_plot(self.train_hist,
os.path.join(self.save_dir, self.dataset,
self.model_name),
self.model_name,
use_subplot=True)
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.loss_plot(self.train_hist,
os.path.join(self.save_dir, self.dataset,
self.model_name),
self.model_name,
use_subplot=True)
def train(self):
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist["per_epoch_time"] = []
self.train_hist['total_time'] = []
if self.gpu_mode:
self.y_real_, self.y_fake_ = Variable(
torch.ones(self.batch_size, 1).cuda()), Variable(
torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(
torch.ones(self.batch_size,
1)), Variable(torch.zeros(self.batch_size, 1))
# self.D.tarin()
print('train start!!')
start_time = time.time()
for epoch in range(self.epoch):
# self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__(
) // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = Variable(x_.cuda()), Variable(z_.cuda())
else:
x_, z_ = Variable(x_), Variable(z_)
"""Update D network"""
self.D_optimizer.zero_grad()
# train with real images
y_hat_real = self.D(x_) # forward pass
D_real_loss = self.BCE_loss(y_hat_real, self.y_real_)
generated_images_ = self.G(z_)
y_hat_fake = self.D(generated_images_)
D_fake_loss = self.BCE_loss(y_hat_fake, self.y_fake_)
D_loss = D_fake_loss + D_real_loss
self.train_hist['D_loss'].append(D_loss.data[0])
D_loss.backward()
self.D_optimizer.step()
"""Update generator network"""
self.G_optimizer.zero_grad()
generated_images_ = self.G(z_)
y_hat_fake = self.D(generated_images_)
G_loss = self.BCE_loss(y_hat_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1),
(iter + 1), self.data_loader.dataset.__len__() //
self.batch_size, D_loss.data[0], G_loss.data[0]))
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
self.visualize_results((epoch + 1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(
self.result_dir + '/' + self.dataset + '/' + self.model_name +
'/' + self.model_name, self.epoch)
utils.loss_plot(
self.train_hist,
os.path.join(self.save_dir, self.dataset, self.model_name),
self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
# print('train at LSGAN')
# self.y_real_, self.y_fake_ = torch.ones(self.batch_size, 1), torch.zeros(self.batch_size, 1)
self.y_real_, self.y_fake_ = torch.zeros(self.batch_size,
1), torch.ones(
self.batch_size, 1)
if self.gpu_mode:
self.y_real_, self.y_fake_ = self.y_real_.cuda(
), self.y_fake_.cuda()
self.D.train()
print('{} training start!!,epoch:{},module stored at:{}'.format(
self.model_name, self.epoch, self.dataset))
start_time = time.time()
# url = os.path.join(self.save_dir, self.dataset, self.model_name)
for epoch in range(self.epoch):
# if epoch == 105:
# self.G = torch.load(os.path.join(url,'LSGAN_105_G.pkl'))
# self.D = torch.load(os.path.join(url,'LSGAN_105_D.pkl'))
# print('reload success!','*'*40)
self.G.train()
try:
if epoch >= 15:
self.G_optimizer.param_groups[0]['lr'] = 0.00009
self.D_optimizer.param_groups[0]['lr'] = 0.00009
elif epoch >= 40:
self.G_optimizer.param_groups[0]['lr'] = 0.00001
self.D_optimizer.param_groups[0]['lr'] = 0.00001
elif epoch >= 70:
self.G_optimizer.param_groups[0]['lr'] = 0.000009
self.D_optimizer.param_groups[0]['lr'] = 0.000009
elif epoch >= 90:
self.G_optimizer.param_groups[0]['lr'] = 0.000001
self.D_optimizer.param_groups[0]['lr'] = 0.000001
elif epoch >= 110:
self.G_optimizer.param_groups[0]['lr'] = 0.0000001
self.D_optimizer.param_groups[0]['lr'] = 0.0000001
except:
# print('1',self.G.__getattribute__)
# print('2',self.G._parameters.__class__,self.G._parameters.__getattribute__)
# try:
# print('3',self.G.param_groups.keys())
# except:
# pass
print('error arise for param_groups at train part')
epoch_start_time = time.time()
# for iter, (x_, _) in enumerate(self.data_loader):
for iter, x_, in enumerate(self.data_loader):
x_ = x_[0]
if iter == self.data_loader.dataset.__len__(
) // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = x_.cuda(), z_.cuda()
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = self.MSE_loss(D_real, self.y_real_)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = self.MSE_loss(D_fake, self.y_fake_)
D_loss = D_real_loss + D_fake_loss
self.train_hist['D_loss'].append(D_loss.item())
D_loss.backward()
self.D_optimizer.step()
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = self.MSE_loss(D_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.item())
G_loss.backward()
self.G_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1),
(iter + 1), self.data_loader.dataset.__len__() //
self.batch_size, D_loss.item(), G_loss.item()),
end=',')
print('lr:%7f' % self.G_optimizer.param_groups[0]['lr'])
self.writer.add_scalar('G_loss', G_loss.item(), self.X)
# writer.add_scalar('G_loss', -G_loss_D, X)
self.writer.add_scalar('D_loss', D_loss.item(), self.X)
self.writer.add_scalars('cross loss', {
'G_loss': D_loss.item(),
'D_loss': D_loss.item()
}, self.X)
self.train_hist['per_epoch_time'].append(time.time() -
epoch_start_time)
# with torch.no_grad():
# self.visualize_results((epoch+1))
if epoch % 5 == 0:
self.load_interval(epoch)
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" %
(np.mean(self.train_hist['per_epoch_time']), self.epoch,
self.train_hist['total_time'][0]))
print("Training finish!... save training results")
save_dir = os.path.join(self.save_dir, self.dataset, self.model_name)
with open(os.path.join(save_dir, self.model_name + '_train_hist.json'),
"a") as f:
json.dump(self.train_hist, f)
self.writer.export_scalars_to_json(
os.path.join(save_dir, self.model_name + '.json'))
self.writer.close()
self.load_interval(epoch)
# self.save()
# utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
# self.epoch)
utils.loss_plot(
self.train_hist,
os.path.join(self.save_dir, self.dataset, self.model_name),
self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['info_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
self.y_real_, self.y_fake_ = torch.ones(self.batch_size, 1), torch.zeros(self.batch_size, 1)
self.y_real_, self.y_fake_ = self.y_real_.cpu(), self.y_fake_.cpu()
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, y_) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__() // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.SUPERVISED == True:
y_disc_ = torch.zeros((self.batch_size, self.len_discrete_code)).scatter_(1, y_.type(torch.LongTensor).unsqueeze(1), 1)
else:
y_disc_ = torch.from_numpy(
np.random.multinomial(1, self.len_discrete_code * [float(1.0 / self.len_discrete_code)],
size=[self.batch_size])).type(torch.FloatTensor)
y_cont_ = torch.from_numpy(np.random.uniform(-1, 1, size=(self.batch_size, 2))).type(torch.FloatTensor)
x_, z_, y_disc_, y_cont_ = x_.cpu(), z_.cpu(), y_disc_.cpu(), y_cont_.cpu()
# update D network
self.D_optimizer.zero_grad()
D_real, _, _ = self.D(x_)
D_real_loss = self.BCE_loss(D_real, self.y_real_)
G_ = self.G(z_, y_cont_, y_disc_)
D_fake, _, _ = self.D(G_)
D_fake_loss = self.BCE_loss(D_fake, self.y_fake_)
D_loss = D_real_loss + D_fake_loss
self.train_hist['D_loss'].append(D_loss.item())
D_loss.backward(retain_graph=True)
self.D_optimizer.step()
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_, y_cont_, y_disc_)
D_fake, D_cont, D_disc = self.D(G_)
G_loss = self.BCE_loss(D_fake, self.y_real_)
self.train_hist['G_loss'].append(G_loss.item())
G_loss.backward(retain_graph=True)
self.G_optimizer.step()
# information loss
disc_loss = self.CE_loss(D_disc, torch.max(y_disc_, 1)[1])
cont_loss = self.MSE_loss(D_cont, y_cont_)
info_loss = disc_loss + cont_loss
self.train_hist['info_loss'].append(info_loss.item())
info_loss.backward()
self.info_optimizer.step()
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f, info_loss: %.8f" %
((epoch + 1), (iter + 1), self.data_loader.dataset.__len__() // self.batch_size, D_loss.item(), G_loss.item(), info_loss.item()))
with torch.no_grad():
self.visualize_results((epoch + 1), (iter + 1))
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name + '_cont',
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)
def train(self):
self.train_hist = {}
self.train_hist['D_loss'] = []
self.train_hist['G_loss'] = []
self.train_hist['per_epoch_time'] = []
self.train_hist['total_time'] = []
if self.gpu_mode:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1).cuda()), Variable(torch.zeros(self.batch_size, 1).cuda())
else:
self.y_real_, self.y_fake_ = Variable(torch.ones(self.batch_size, 1)), Variable(torch.zeros(self.batch_size, 1))
self.D.train()
print('training start!!')
start_time = time.time()
for epoch in range(self.epoch):
self.G.train()
epoch_start_time = time.time()
for iter, (x_, _) in enumerate(self.data_loader):
if iter == self.data_loader.dataset.__len__() // self.batch_size:
break
z_ = torch.rand((self.batch_size, self.z_dim))
if self.gpu_mode:
x_, z_ = Variable(x_.cuda()), Variable(z_.cuda())
else:
x_, z_ = Variable(x_), Variable(z_)
# update D network
self.D_optimizer.zero_grad()
D_real = self.D(x_)
D_real_loss = -torch.mean(D_real)
G_ = self.G(z_)
D_fake = self.D(G_)
D_fake_loss = torch.mean(D_fake)
# gradient penalty
if self.gpu_mode:
alpha = torch.rand(x_.size()).cuda()
else:
alpha = torch.rand(x_.size())
x_hat = Variable(alpha * x_.data + (1 - alpha) * G_.data, requires_grad=True)
pred_hat = self.D(x_hat)
if self.gpu_mode:
gradients = grad(outputs=pred_hat, inputs=x_hat, grad_outputs=torch.ones(pred_hat.size()).cuda(),
create_graph=True, retain_graph=True, only_inputs=True)[0]
else:
gradients = grad(outputs=pred_hat, inputs=x_hat, grad_outputs=torch.ones(pred_hat.size()),
create_graph=True, retain_graph=True, only_inputs=True)[0]
gradient_penalty = self.lambda_ * ((gradients.view(gradients.size()[0], -1).norm(2, 1) - 1) ** 2).mean()
D_loss = D_real_loss + D_fake_loss + gradient_penalty
D_loss.backward()
self.D_optimizer.step()
if ((iter+1) % self.n_critic) == 0:
# update G network
self.G_optimizer.zero_grad()
G_ = self.G(z_)
D_fake = self.D(G_)
G_loss = -torch.mean(D_fake)
self.train_hist['G_loss'].append(G_loss.data[0])
G_loss.backward()
self.G_optimizer.step()
self.train_hist['D_loss'].append(D_loss.data[0])
if ((iter + 1) % 100) == 0:
print("Epoch: [%2d] [%4d/%4d] D_loss: %.8f, G_loss: %.8f" %
((epoch + 1), (iter + 1), self.data_loader.dataset.__len__() // self.batch_size, D_loss.data[0], G_loss.data[0]))
self.train_hist['per_epoch_time'].append(time.time() - epoch_start_time)
self.visualize_results((epoch+1))
self.train_hist['total_time'].append(time.time() - start_time)
print("Avg one epoch time: %.2f, total %d epochs time: %.2f" % (np.mean(self.train_hist['per_epoch_time']),
self.epoch, self.train_hist['total_time'][0]))
print("Training finish!... save training results")
self.save()
utils.generate_animation(self.result_dir + '/' + self.dataset + '/' + self.model_name + '/' + self.model_name,
self.epoch)
utils.loss_plot(self.train_hist, os.path.join(self.save_dir, self.dataset, self.model_name), self.model_name)