Normalize step size to the batch size used - #431
Merged
Conversation
Closed
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
conradsnicta
requested changes
Sep 2, 2025
conradsnicta
approved these changes
Sep 3, 2025
Member
Author
|
Note that this will also need a downstream patch in mlpack to work correctly. I'll work on that next, although it only becomes fully necessary when we release ensmallen 3.x. |
Member
Author
|
The mlpack integration test failure is expected---once mlpack/mlpack#4002 is merged, that will be fixed. |
This was referenced Sep 23, 2025
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This handles #428. I went through all separable differentiable optimizers and found that all of them that use
SGDinternally do not normalize by the batch size when taking a step. This problem also exists withEve. But other separable optimizers do not have this issue because they are techniques that specifically account for the batch size (likeBigBatchSGD, orSVRG, or others).This is a breaking change, so it will necessitate a major release. I also added the
ENS_OLD_SEPARABLE_STEP_BEHAVIORmacro to allow users to continue to use the old behavior so that it doesn't break current workflows. I could be persuaded to remove that, but I thought it was a nice thing to do.