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.

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

Use PackedSequence for ConvRNN #126

Merged
merged 7 commits into from Jun 24, 2018

Conversation

Projects
None yet
3 participants
@daemon
Copy link
Member

commented Jun 23, 2018

Zero-padding is technically incompatible with the recurrent layers, since the hidden outputs would still be altered. Using PackedSequence is the correct way to deal with this issue. Thus, I've invalidated the past results and will report the current ones as soon as the model finishes training.

@daemon daemon requested review from tuzhucheng and Victor0118 Jun 23, 2018

@@ -3,7 +3,7 @@
import random

from torch import utils
from torch.optim.lr_scheduler import ReduceLROnPlateau
from tqdm import tqdm

This comment has been minimized.

Copy link
@tuzhucheng

tuzhucheng Jun 24, 2018

Member

Be sure to add tqdm to requirements.txt

@tuzhucheng

This comment has been minimized.

Copy link
Member

commented Jun 24, 2018

By the way please migrate to SST1 in the common datasets module when you get a chance?

@Victor0118

This comment has been minimized.

Copy link
Member

commented Jun 24, 2018

LGTM. Might merge after you push some reasonable results to ensure it works well.

@daemon

This comment has been minimized.

Copy link
Member Author

commented Jun 24, 2018

@tuzhucheng Sounds good, I'll do that in a future PR.

@daemon

This comment has been minimized.

Copy link
Member Author

commented Jun 24, 2018

Also, I've added results using the default hyperparameters specified in the paper. They're around 2.7 points off.

@daemon daemon merged commit e4149ff into master Jun 24, 2018

@daemon daemon deleted the daemon/convrnn-fix branch Jun 24, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.