def forward(self, source, target):
source = self.sharedNet(source)
loss = 0
if self.training == True:
target = self.sharedNet(target)
loss = mmd.mmd_linear(source, target)
source = self.cls_fc(source)
return source, loss
def forward(self, source, target):
source = self.sharedNet(source)
loss = 0
if self.training == True:
target = self.sharedNet(target)
loss = mmd.mmd_linear(source, target)
source = self.cls_fc(source)
return source, loss
def forward(self, source, target):
source = self.features(source)
source = source.view(source.size(0), -1)
source = self.classifier(source)
source = self.bottleneck(source)
mmd_loss = 0
if self.training:
target = self.features(target)
target = target.view(target.size(0), -1)
target = self.classifier(target)
target = self.bottleneck(target)
mmd_loss += mmd.mmd_linear(source, target)
result = self.final_classifier(source)
return result, mmd_loss
def compute_adapt_loss(self, X, Y, adapt_loss):
"""Compute adaptation loss, currently we support mmd and coral
Arguments:
X {tensor} -- source matrix
Y {tensor} -- target matrix
adapt_loss {string} -- loss type, 'mmd' or 'coral'. You can add your own loss
Returns:
[tensor] -- adaptation loss tensor
"""
if adapt_loss == 'mmd':
loss = mmd.mmd_linear(X, Y)
elif adapt_loss == 'coral':
loss = CORAL(X, Y)
else:
loss = 0
return loss
def train(model, config, epoch):
model.class_classifier.train()
model.feature.train()
iter_source = iter(config['source_train_loader'])
iter_target = iter(config['target_train_loader'])
len_source_loader = len(config['source_train_loader'])
len_target_loader = len(config['target_train_loader'])
num_iter = len_source_loader
for i in range(1, num_iter):
data_source, label_source = iter_source.next()
data_target, _ = iter_target.next()
if i % len_target_loader == 0:
iter_target = iter(config['target_train_loader'])
if torch.cuda.is_available():
data_source, label_source = data_source.cuda(
), label_source.cuda()
data_target = data_target.cuda()
optimizer.zero_grad()
preds = model.class_classify(data_source)
loss_cls = criterion(preds, label_source)
source = model.feature(data_source)
source = source.view(source.size(0), -1)
target = model.feature(data_target)
target = target.view(target.size(0), -1)
loss_mmd = mmd.mmd_linear(source, target)
loss = loss_cls + config['mmd_gamma'] * loss_mmd
if i % 50 == 0:
print(
'loss_cls {}, loss_mmd {}, gamma {}, total loss {}'.format(
loss_cls.item(), loss_mmd.item(), config['mmd_gamma'],
loss.item()))
loss.backward()
optimizer.step()