Drop incomplete batches for Ray and Pandas to prevent Batchnorm computation errors #2778
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Outdated suggestions cannot be applied.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 PR implements logic to drop the last batch for Pandas and Ray dataset batchers to prevent issues when there is only 1 row in a batch.
A single row in a batch causes issues when computing Batchnorm (such as in FC layers, Tabnet, etc.) because it can't compute the mean/stddev for a single sample. Additionally, Ludwig has some logic for Tabnet that has a conditional check for sample_size == 1 that we can now safely remove.
Skipping adding tests since our current test suite already has tests that use the batcher classes for model training, so if the tests pass, then this logic should work as well. I modified individual tests locally to be sure.
Note: This only drops incomplete training batches if the batch size < total number of rows in the dataset. Additionally, it will always keep incomplete training batches for validation/test sets.
To follow: