Skip to content

v0.13.0

Choose a tag to compare

View all tags
@francois-drielsma francois-drielsma released this 31 May 08:25
· 371 commits to main since this release
v0.13.0
2286c5c

Added

  • Visualization docs: Add a dedicated spine.vis README covering the reorganized trace, drawer, metric, and layout structure introduced in #131.
  • Driver logging backends: Add a structured LogManager with optional TensorBoard integration and CSV/timing/memory logging support as part of the driver refactor in #132.
  • Index-span metadata: Add explicit parser payload classes and span-aware batch metadata for flat indexes, index lists, and edge indexes in #133.

Changed

  • Math package cleanup: Review and tighten spine.math typing, tests, and helper behavior, including the iterative pair-distance path now used by the full-chain regression baseline in #130.
  • Visualization package structure: Reorganize spine.vis into explicit trace, drawer, metric, and layout subpackages, preserve direct import exposure through spine.vis, and restore comprehensive coverage in #131.
  • Driver and I/O ownership boundaries: Refactor driver initialization, move batching/unwrapping responsibilities under spine.io, introduce an IOManager, and separate structured logging concerns through #132.
  • Index batching model: Replace implicit global index shifts with explicit per-entry spans throughout parsing, collation, overlay, unwrapping, and cached HDF5 index handling in #133.
  • Container/runtime defaults: Simplify container metadata handling and restore multi-rank training summaries so distributed runs emit one coherent per-rank progress table from the main process.

Fixed

  • Closest/farthest pair utilities: Fix pair-distance helper behavior used by GrapPA feature engineering and align the deterministic full-chain regression reference with the iterative implementation in #130.
  • Visualization regressions: Fix CI regressions in the reorganized output drawer, restore coverage, and preserve existing behavior after the package shuffle in #131.
  • CLI and runtime polish: Move banner printing to the CLI, reduce duplicate startup output, improve bin-package typing, and expand coverage around the main runtime and entrypoint helpers as part of #132.
  • Index list semantics: Preserve list-backed IndexBatch behavior for object-array-backed cluster lists, fix downstream PPN cluster access after the span refactor, and require count metadata for HDF5 index parsers where spans must be reconstructed after #133.
  • Distributed training summaries: Gather per-rank iteration rows onto rank 0 so training logs once again report timing, memory, loss, and accuracy for every process without duplicating the full header block.

Full Changelog: v0.12.4...v0.13.0

Assets 2
Loading