Anserini Release Notes (v0.39.0)

Release date: January 12, 2025
Lucene version: Lucene 9.9.1

Summary of Changes

Added support for indexing dense vectors in Parquet format:
- Created ParquetDenseVectorCollection and ParquetDenseVectorDocumentGenerator.
- Used strategicblue/parquet-floor dependency.
- Refactored DenseVectorDocumentGenerator to JsonDenseVectorDocumentGenerator.
- Refactored InvertedDenseVectorDocumentGenerator to JsonInvertedDenseVectorDocumentGenerator.
- Established parallel class structure for jsonl and Parquet formats.
- Added dynamic type support (float/double) for ParquetDenseVectorDocumentGenerator.
Added bindings: RAG24 topics with Arctic emeddings, RAG24 UMBRELA qrels.
Added initial fusion implementation: regressions script w/ initial yaml config.
Installed new regressions: RAG24 baselines, MS MARCO Passage V1 Parquet, BEIR Parquet.
Refactored approach to tolerance checking and tweaked yaml configs.
Improved search UI landing page and search bar when displaying results.
Renamed cos-dpr to cos-dpr: refactored prebuilt indexes, symbols etc.

Sorted by number of commits:

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