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

Information regarding training time #13

Open
backpropper opened this issue Oct 16, 2017 · 6 comments
Open

Information regarding training time #13

backpropper opened this issue Oct 16, 2017 · 6 comments

Comments

@backpropper
Copy link
Contributor

backpropper commented Oct 16, 2017

Can you please provide some information regarding the training time (with the hardware specifications) and the size of the dataset used?

Thanks!

@Cadene
Copy link
Owner

Cadene commented Oct 16, 2017

On one TitanX Pascal, on VQA2, when data loading time is closed to 0, you can expect 5 hours before convergence for a NoAtt model and 12 hours for a Att model.

@backpropper
Copy link
Contributor Author

Same for VQA1 too?

@Cadene
Copy link
Owner

Cadene commented Oct 16, 2017

I don't remember for VQA1, but as it is composed of half of the questions, I would say 3 hours for NoAtt and 7 hours for Att. Let me know :P

@backpropper
Copy link
Contributor Author

backpropper commented Oct 16, 2017

And you used the whole training set for training 100 epochs?

@Cadene
Copy link
Owner

Cadene commented Oct 16, 2017

train/val or trainval/test
But in both cases, always the full dataset

@backpropper
Copy link
Contributor Author

backpropper commented Oct 17, 2017

But you only train with the examples whose multiple choice answer is in the top n% (n determined by nans) and discard other examples. Is that correct?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants