Skip to content

v0.3.2

Choose a tag to compare

@ScionKim ScionKim released this 04 Sep 06:46
· 110 commits to main since this release
36822b5

Performance

  • Increase the internal batch-sizing budget for donor_policy="available" from 16 MiB to 128 MiB to reduce small-batch overhead on larger training datasets.
  • Leave complete-donor processing, constructor parameters, and thread-count configuration unchanged.

Memory and performance notes

  • Larger batches can increase peak memory usage. The 128 MiB batching budget is not a total process-memory limit.
  • Speed and memory usage depend on data and hardware; improvements over KNNImputer are not guaranteed.

Tests and documentation

  • Add regression checks for default batching, numerical edge cases, and output consistency.
  • Preserve the historical 16 MiB benchmark baseline.
  • Streamline the README and document benchmark results and memory tradeoffs.

Benchmark report