Skip to content

CHANGELOG.md

Latest commit

 

History

History
153 lines (94 loc) · 3.74 KB

CHANGELOG.md

File metadata and controls

153 lines (94 loc) · 3.74 KB
Raw

CHANGELOG

This is a changelog of all notable changes to this project. We adhere to Semantic Versioning.

Unreleased

Added

  • AttributeToClassTransform, AttributeToRegressionTransform and NameToClass3MTransform (with NAME_TO_CLASS_3M): stateless task transforms whose targets are fixed functions of the stored attributes.

    MANTRADivided for tuning the on-the-fly balancing (target_count, n_moves, use_topology_changes, max_vertices, verbose).

Removed

  • CreateLabels, which assigned class indices in encounter order and therefore depended on dataset traversal; use AttributeToClassTransform or a name transform instead.

Changed

  • balanced=True now computes the balanced dataset during process() via Pachner-move augmentation and deduplication instead of downloading a pre-generated release asset. This also fixes the 404 for recent releases, which no longer shipped balanced assets.

  • balance_dataset draws augmentations only from original entries (never from augmented copies), keeps a random subsample per class instead of the smallest triangulations, deduplicates only classes that gained augmented entries, and raises informative ValueErrors instead of bare assertions.

  • Split caches now also encode split_proportions and stratified, so changing either re-processes instead of silently serving stale splits.

  • The exact-duplicates warning for oversampled OOD classes now also covers stellar subdivision with fraction=1.0, which is just as deterministic as barycentric subdivision.

Removed

  • scripts/generate_balanced.py (superseded by on-the-fly balancing) and the balanced parameter of the internal dataset URL helper.

v0.0.16

Added

  • Canonical names for all 2-manifolds, thus adding almost 40000 new names.

Fixed

  • Renamed vertex-transitive attribute to vertex_transitive.

  • Fixed generation of release notes.

v0.0.15

Added

  • New deduplication routine for merging different types of manifold datasets.

  • Added 4787 triangulations of small valence, originally collected by Frank Lutz but published without homology group information. This information was added to the upstream repository and is now part of the larger 3-manifold dataset.

v0.0.14

Added

  • Proper handling of the dataset versions. In case the provided version does not exist a ValueError will be raised and all available versions will be listed to the user.

Fixed

  • Changed root path to ensure the 2D and 3D datasets do not resolve to the same file.

v0.0.13

Fixed

  • Addresses some issues with building the documentation.

v0.0.12

Added

  • Included transforms in main documentation.
  • Extended pyproject.toml with project URLs.
  • Fixed look-and-feel of PyPI documentation.

v0.0.11

Fixed

  • Minor changes to documentation of individual transforms.
  • Simplified dependencies.

v0.0.10

Fixed

  • Minor changes to the release scripts.

v0.0.9

Fixed

  • Finally using the correct tag by creating a new release before we send everything to Zenodo. Life is hard.

v0.0.8

Fixed

  • Using proper logic to name releases on Zenodo. This is just a minor choice of aesthetics.

v0.0.7

Added

  • Depositing the generated files automatically on Zenodo, thus adding additional versioning and the ability to refer to a particular version using a DOI.

v0.0.6

Fixed

  • Using better class names to be consistent with the paper.

v0.0.5

Added

  • Added automatic upload of the changelog to the GitHub release.

v0.0.4

Fixed

  • Fixed a tag that prevented fetching the last version of the dataset.
  • Added synchronization between the released dataset version and package version on PyPi.