Anserini Release Notes (v0.24.0)

Release date: December 28, 2023
Lucene version: Lucene 9.9.1

Summary of Changes

Upgraded to Lucene 9.9.1:
- Refactored code to use latest codecs.
- Added regressions to take advantage of HNSW int8 quantization (but doesn't work for OpenAI Ada2).
Major refactoring of search code paths:
- Refactored to create HnswDenseSearcher and InvertedDenseSearcher to provide Python bindings.
- Refactored SearchHnswDenseVectors and SearchInvertedDenseVectors as wrappers to provide main entry points to above searchers.
- Improved alignment between SearchCollection and dense vector search classes above, extracting common code into helper classes.
- Aligned ScoredDoc and ScoredDocs (was previously ScoredDocuments) as container objects for Lucene results.
- Refactored searchers to use ScoredDoc instead of class-specific Result objects.
- Imposed uniform camelCasing in search args.
Major refactoring of indexing pipelines:
- Extracted common code paths from IndexCollection, IndexHnswDenseVectors, and IndexInvertedDenseVectors into AbstractIndexer.
- Imposed uniform camelCasing in index args.
Improved test coverage.
Added basic ability to download pre-built indexes.

Sorted by number of commits:

All contributors with five or more commits, sorted by number of commits, according to GitHub: