Skip to content

fix(v1): honor batch size in train_gen without permutation#391

Open
kuishou68 wants to merge 1 commit intogoogle-research:masterfrom
kuishou68:cocoon/fix-train-gen-batching
Open

fix(v1): honor batch size in train_gen without permutation#391
kuishou68 wants to merge 1 commit intogoogle-research:masterfrom
kuishou68:cocoon/fix-train-gen-batching

Conversation

@kuishou68
Copy link
Copy Markdown

@kuishou68 kuishou68 commented Apr 4, 2026

Summary

  • keep train_gen() batches aligned with batch_size when permute=False
  • mirror the deterministic chunking behavior already used by test_val_gen()
  • add a regression test covering the non-permuted batching path

Notes

  • I could not run the Python test suite in this sandbox because the available Python runtime is incomplete, so this patch was validated by code-path review and a focused regression test addition.

Closes #359

@kuishou68 kuishou68 mentioned this pull request Apr 4, 2026
Open
darkpowerxo added a commit to darkpowerxo/timesfm that referenced this pull request Apr 8, 2026
@darkpowerxo
fix: respect batch_size in v1 data_loader when permute=False
fb8ac51 
Apply changes from PR google-research#391 by @MarcoGorworworelli:
- Fix train_gen() to iterate in proper batch_size chunks instead of
  yielding all time series at once when permute=False
- Add test_data_loader.py to verify batch boundaries

Co-authored-by: MarcoGorelli <marco_gorelli@protonmail.com>
@darkpowerxo darkpowerxo mentioned this pull request Apr 8, 2026
Closed
darkpowerxo added a commit to darkpowerxo/timesfm that referenced this pull request Apr 8, 2026
@darkpowerxo
fix: respect batch_size in v1 data_loader when permute=False
07bf9e8 
Apply changes from PR google-research#391 by @MarcoGorworworelli:
- Fix train_gen() to iterate in proper batch_size chunks instead of
  yielding all time series at once when permute=False
- Add test_data_loader.py to verify batch boundaries
@darkpowerxo darkpowerxo mentioned this pull request Apr 8, 2026
Merged
darkpowerxo added a commit to darkpowerxo/timesfm that referenced this pull request Apr 8, 2026
@darkpowerxo
fix: respect batch_size in v1 data_loader when permute=False
1bb44d5 
Apply changes from PR google-research#391 by @MarcoGorworworelli:
- Fix train_gen() to iterate in proper batch_size chunks instead of
  yielding all time series at once when permute=False
- Add test_data_loader.py to verify batch boundaries
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

train_gen loads indices of all time series instead of batch-size if self.permute is False

1 participant

@kuishou68