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

No dataset constants for "femnist" dataset #1

Open
meryemJanatiIdrissi opened this issue Oct 27, 2020 · 1 comment
Open

No dataset constants for "femnist" dataset #1

meryemJanatiIdrissi opened this issue Oct 27, 2020 · 1 comment

Comments

@meryemJanatiIdrissi
Copy link

when i run the experiments for femnist dataset i run to this error:

FileNotFoundError: [Errno 2] No such file or directory: 'dataset_constants/femnist_train_mean_std'

and other errors in the function update
update(self.updates, aggregation, max_update_norm=MAX_UPDATE_NORM)
specifically the "max_update_norm" parameter.

I tried to ignore these errors by not doing the normalization and deleting the "max_update_norm" parameter however the accuracy is very bad (0.05)

do you please have a solution for this?

@krishnap25
Copy link
Owner

Hello,

Sorry for the late response. We updated the dataset_constants folder with the EMNIST mean and std values. The update function works fine for me. Could you please post the full error message?

More details:

  • The normalization allows for larger learning rates, leading to faster convergence. You can run it without the normalization, but you'll have to use a smaller learning rate, such as 0.05, for instance (with normalization, we use a learning rate of 1.0).

  • The max_update_norm is just a safety check to ensure that a large unexpected update does not cause divergence. You do not need it to reproduce the experiments reported in our paper (i.e., you can safely set it to None).

Please let us know if you need anything else.

Best,
Krishna

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