Repository navigation
Highlights
- Added
--num_workersand matching model API options, defaulting to single-process data loading for more reliable cross-platform inference. Thanks @muschellij2 for #71. - Added
--sample_ratefor CSV and pickle inputs while preserving automatic rate inference by default. Thanks @muschellij2 for #73. - Added
AsleepVersionto preprocessing metadata, including refreshed metadata for cached outputs. Thanks @muschellij2 for #74. - Restored day-level and aggregate summary generation from persisted timestamps.
- Improved PyTorch and NumPy compatibility across inference, sequence handling, missing labels, device paths, and parallel-model checkpoint serialization.
- Added packaged typing metadata for downstream type checkers.
- Added a documented release process and corrected publishing automation to run from published GitHub Releases.
Requirements
- NumPy 1.24 or newer is now required.
Full changelog: 0.4.18...0.5.0