Skip to content

v0.3.6

Choose a tag to compare

@ScionKim ScionKim released this 06 Sep 14:22
· 96 commits to main since this release
da51cdd

Fix support for NumPy integer n_neighbors values, including parameter grids generated with np.arange() in Pipeline and GridSearchCV.

  • Accept positive Python and NumPy integer scalars for both donor policies.
  • Preserve the original estimator parameter for scikit-learn cloning.
  • Convert neighbor counts to Python integers before FAISS search and candidate-expansion arithmetic.
  • Explicitly reject Python and NumPy booleans. Python True was previously accepted as 1.
  • Preserve existing donor-count limits and failed-fit cleanup.

Validation: 189 tests passed for the fix, including 39 new regression cases covering NumPy-generated grids, signed and unsigned integers, cloning, input preservation, invalid parameters, donor limits, and candidate-expansion overflow.