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

Jump to bottom

Test and Travis-CI #20

Merged
merged 18 commits into from
Dec 22, 2017
Merged

Test and Travis-CI #20

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
c0d9744
add travis test
ShigekiKarita Dec 18, 2017
23f5696
fix .travis.yml
ShigekiKarita Dec 18, 2017
a7a43f3
make pytorch initialization consistent to chainer and test partialy
Dec 19, 2017
2462ce3
fix encoding utf-8
Dec 19, 2017
d3f6bb1
more compatible chainer/pytorch
Dec 19, 2017
fe60ae8
remove unused main
Dec 19, 2017
f08c911
make accuracy consistent
Dec 19, 2017
4d48754
add more test
Dec 20, 2017
62a1cb0
add forgotten tanh in encoder, chainer-like ctc and test them
ShigekiKarita Dec 21, 2017
fd4dac7
Merge remote-tracking branch 'espnet/master' into test-pytorch-loss
Dec 21, 2017
517c95e
add more tests on optim / grad
ShigekiKarita Dec 21, 2017
6b0b4a8
add more tests on grad
ShigekiKarita Dec 21, 2017
68f2c88
workaround for zero-length
ShigekiKarita Dec 21, 2017
d2fdf0a
fix pytest to import torch
ShigekiKarita Dec 21, 2017
11c14d4
fix .travis.yml
ShigekiKarita Dec 21, 2017
696a3cc
reflect reviews from kan-bayashi
ShigekiKarita Dec 22, 2017
b4129da
revert attloc tanh
ShigekiKarita Dec 22, 2017
e3a19ab
update Makefile
ShigekiKarita Dec 22, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
13 changes: 7 additions & 6 deletions src/nets/e2e_asr_attctc_th.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -408,7 +408,7 @@ def forward(self, enc_hs_pad, enc_hs_len, dec_z, att_prev, scaling=2.0):
self.enc_h = enc_hs_pad # utt x frame x hdim
self.h_length = self.enc_h.shape[1]
# utt x frame x att_dim
self.pre_compute_enc_h = linear_tensor(self.mlp_enc, self.enc_h)
self.pre_compute_enc_h = torch.tanh(linear_tensor(self.mlp_enc, self.enc_h))

if dec_z is None:
dec_z = Variable(enc_hs_pad.data.new(batch, self.dunits).zero_())
Expand Down Expand Up @@ -469,7 +469,7 @@ def forward(self, enc_hs_pad, enc_hs_len, dec_z, att_prev, scaling=2.0):
self.enc_h = enc_hs_pad # utt x frame x hdim
self.h_length = self.enc_h.shape[1]
# utt x frame x att_dim
self.pre_compute_enc_h = linear_tensor(self.mlp_enc, self.enc_h)
self.pre_compute_enc_h = torch.tanh(linear_tensor(self.mlp_enc, self.enc_h))
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

tanh op is performed for only AttDot in Chainer.
I think it is not needed for AttLoc.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

oh that is right


if dec_z is None:
dec_z = Variable(enc_hs_pad.data.new(batch, self.dunits).zero_())
Expand Down Expand Up @@ -509,10 +509,11 @@ def forward(self, enc_hs_pad, enc_hs_len, dec_z, att_prev, scaling=2.0):


def th_accuracy(y_all, pad_target, ignore_label):
acc = 0
pad_pred = y_all.data.view(pad_target.size(0), pad_target.size(1), y_all.size(1)).max(2)[1]
mask = pad_target.data != ignore_label
return torch.sum(pad_pred.masked_select(mask) == pad_target.data.masked_select(mask)) / torch.sum(mask)
numerator = torch.sum(pad_pred.masked_select(mask) == pad_target.data.masked_select(mask))
denominator = torch.sum(mask)
return float(numerator) / float(denominator)


# ------------- Decoder Network ----------------------------------------------------------------------------------------
Expand Down Expand Up @@ -601,8 +602,8 @@ def forward(self, hpad, hlen, ys):
for (i, y_hat_), y_true_ in zip(enumerate(y_hat.data.cpu().numpy()), y_true.data.cpu().numpy()):
if i == MAX_DECODER_OUTPUT:
break
idx_hat = np.argmax(y_hat_[y_true_ != -1], axis=1)
idx_true = y_true_[y_true_ != -1]
idx_hat = np.argmax(y_hat_[y_true_ != self.ignore_id], axis=1)
idx_true = y_true_[y_true_ != self.ignore_id]
seq_hat = [self.char_list[int(idx)] for idx in idx_hat]
seq_true = [self.char_list[int(idx)] for idx in idx_true]
seq_hat = "".join(seq_hat)
Expand Down
1 change: 1 addition & 0 deletions test/test_torch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
import pytest
pytest.importorskip('torch')
import torch
from torch.autograd import Variable
from e2e_asr_attctc_th import pad_list, mask_by_length


Expand Down