def train(self):
for epoch in range(10):
for it, (x, y) in enumerate(self.data_loader):
self.optim.zero_grad()
x = torch.bernoulli(x)
if cuda:
x = x.cuda()
x = Variable(x.view(-1, 1, 28, 28))
out = nn_.sigmoid(self.mdl((x, 0))[0]).permute(0, 3, 1, 2)
loss = utils.bceloss(out, x).sum(1).sum(1).sum(1).mean()
loss.backward()
self.optim.step()
if ((it + 1) % 100) == 0:
print 'Epoch: [%2d] [%4d/%4d] loss: %.8f' % \
(epoch+1, it+1,
self.data_loader.dataset.__len__() // 32,
loss.data[0])
def train(self):
for epoch in range(10):
for it, (x, y) in enumerate(self.data_loader):
self.optim.zero_grad()
x = torch.bernoulli(x)
x = Variable(x.view(-1, 784))
out = nn_.sigmoid(self.mdl(x)[:,:,0])
loss = utils.bceloss(out, x).sum(1).mean()
loss.backward()
self.optim.step()
if ((it + 1) % 10) == 0:
print 'Epoch: [%2d] [%4d/%4d] loss: %.8f' % \
(epoch+1, it+1,
self.data_loader.dataset.__len__() // 32,
loss.data[0])
self.mdl.randomize()
def evaluate(self, z):
prob = self.sigmoid(self.logits)
z = z * 0.5 + 0.5
return -nn_.sum_from_one(bceloss(prob, z))
def evaluate(self, input, output):
prob = self.forward(input) * 0.5 + 0.5
output = output * 0.5 + 0.5
return -nn_.sum_from_one(bceloss(prob, output))