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

[SYSTEMDS-2550] Batch scaling and weighing of imbalanced workers #1141

Closed

Conversation

tobiasrieger
Copy link
Contributor

This PR provides a number of updates to the Federated Parameter Server

  • Batch Scaling to fix imbalance during runtime
  • Weighing of gradients relative to the data size the represent for all runtime balancing schemes
  • Seeding for data partitioners
  • New test cases and better logging

Copy link
Contributor

@sebwrede sebwrede left a comment

Choose a reason for hiding this comment

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

It generally looks good, I added a few comments.

It could make sense to add documentation to each of the subclasses of DataPartitionFederatedScheme describing the partitioning scheme the class represents. This would make it easier to quickly understand the differences between the partitioning scheme classes, which would make it easier for you to present your work and it would be easier for future development/review of the classes.

@tobiasrieger
Copy link
Contributor Author

@sebwrede I have addressed all other proposed changes in the new commit. Thanks!

@sebwrede
Copy link
Contributor

sebwrede commented Jan 5, 2021

LGTM!

@mboehm7
Copy link
Contributor

mboehm7 commented Jan 9, 2021

LGTM - thanks @tobiasrieger for the patch and @sebwrede for the review. I only modified some formatting issues, reduced the visibility of some members, change the string concatenation, and guarded the info logging (to avoid unnecessary string concatenation).

@asfgit asfgit closed this in 460c394 Jan 9, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
3 participants