Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

clipping before optimizing step. #5

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

clipping before optimizing step. #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4f83c55
update .item, removed the normalization because reconstructiong direc…
Jul 10, 2019
ecdabed
updated model for dealing with nan's
Jul 10, 2019
9689406
removed unwanted documentation
Jul 10, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,12 +86,12 @@ def forward(self, x):
#encoder
enc_t = self.enc(torch.cat([phi_x_t, h[-1]], 1))
enc_mean_t = self.enc_mean(enc_t)
enc_std_t = self.enc_std(enc_t)
enc_std_t = self.enc_std(enc_t) + 1e-5

#prior
prior_t = self.prior(h[-1])
prior_mean_t = self.prior_mean(prior_t)
prior_std_t = self.prior_std(prior_t)
prior_std_t = self.prior_std(prior_t) + 1e-5

#sampling and reparameterization
z_t = self._reparameterized_sample(enc_mean_t, enc_std_t)
Expand All @@ -100,7 +100,7 @@ def forward(self, x):
#decoder
dec_t = self.dec(torch.cat([phi_z_t, h[-1]], 1))
dec_mean_t = self.dec_mean(dec_t)
dec_std_t = self.dec_std(dec_t)
dec_std_t = self.dec_std(dec_t) + 1e-5

#recurrence
_, h = self.rnn(torch.cat([phi_x_t, phi_z_t], 1).unsqueeze(0), h)
Expand All @@ -125,6 +125,7 @@ def sample(self, seq_len):
sample = torch.zeros(seq_len, self.x_dim)

h = Variable(torch.zeros(self.n_layers, 1, self.h_dim))

for t in range(seq_len):

#prior
Expand Down Expand Up @@ -172,7 +173,7 @@ def _kld_gauss(self, mean_1, std_1, mean_2, std_2):

kld_element = (2 * torch.log(std_2) - 2 * torch.log(std_1) +
(std_1.pow(2) + (mean_1 - mean_2).pow(2)) /
std_2.pow(2) - 1)
(std_2).pow(2) - 1)
return 0.5 * torch.sum(kld_element)


Expand Down
28 changes: 15 additions & 13 deletions train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,31 +22,33 @@ def train(epoch):
#data = Variable(data)
#to remove eventually
data = Variable(data.squeeze().transpose(0, 1))
data = (data - data.min().data[0]) / (data.max().data[0] - data.min().data[0])
#data = (data - data.min().item()) / (data.max().item() - data.min().item())

#forward + backward + optimize
optimizer.zero_grad()
kld_loss, nll_loss, _, _ = model(data)
loss = kld_loss + nll_loss
loss.backward()
optimizer.step()

#grad norm clipping, only in pytorch version >= 1.10
nn.utils.clip_grad_norm(model.parameters(), clip)
nn.utils.clip_grad_norm_(model.parameters(), clip)

optimizer.step()


#printing
if batch_idx % print_every == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\t KLD Loss: {:.6f} \t NLL Loss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader),
kld_loss.data[0] / batch_size,
nll_loss.data[0] / batch_size))
kld_loss.item() / batch_size,
nll_loss.item() / batch_size))

sample = model.sample(28)
plt.imshow(sample.numpy())
plt.pause(1e-6)
#sample = model.sample(28)
#plt.imshow(sample.numpy())
#plt.pause(1e-6)

train_loss += loss.data[0]
train_loss += loss.item()


print('====> Epoch: {} Average loss: {:.4f}'.format(
Expand All @@ -62,11 +64,11 @@ def test(epoch):

#data = Variable(data)
data = Variable(data.squeeze().transpose(0, 1))
data = (data - data.min().data[0]) / (data.max().data[0] - data.min().data[0])
#data = (data - data.min().item()) / (data.max().item() - data.min().item())

kld_loss, nll_loss, _, _ = model(data)
mean_kld_loss += kld_loss.data[0]
mean_nll_loss += nll_loss.data[0]
mean_kld_loss += kld_loss.item()
mean_nll_loss += nll_loss.item()

mean_kld_loss /= len(test_loader.dataset)
mean_nll_loss /= len(test_loader.dataset)
Expand Down