v0.15.0: Experimental FTS, packed struct encoding

What's Changed

Breaking Changes 🛠

• feat: upgrade DataFusion, Arrow, PyO3, ObjectStore by @wjones127 in #2594

New Features 🎉

New feature breakdown

• feat: add fast search to to_table method by @chebbyChefNEQ in #2591
• feat: loosen pyarrow pin by @wjones127 in #2597
• feat: add FSST string compression by @broccoliSpicy in #2470
• feat: integrate inverted index into lance index APIs by @BubbleCal in #2577
  • You can now create full text search indices and pass text search queries into the full_text_query argument
• feat: add projection to take and take_rows by @westonpace in #2623
• feat: add storage_options to _BaseLanceDatasink, LanceDatasink, LanceCommitter by @BitPhinix in #2619
• feat: add bitpack encoding for LanceV2 by @albertlockett in #2333
• feat: add a packed struct encoding to lance by @raunaks13 in #2593
• feat: support deletion with stable row id and scalar indices by @wjones127 in #2609

Bug Fixes 🐛

• fix: avoid making empty range requests as these will err on S3 by @westonpace in #2592
• fix: upgrade AWS SDK by @wjones127 in #2616
• fix: specify behavior version when creating dynamodb config by @westonpace in #2624
• fix: pin numpy version to avoid numpy >= 2.0 until pyarrow supports it by @westonpace in #2637

Performance Improvements 🚀

• perf: improve inverted index performance by @BubbleCal in #2574
• perf: improve v2 scan performance by @westonpace in #2604
• perf: allow fragment scan for nearest query if there is a prefilter by @jiachengdb in #2631
• perf: speed up fsst decompression by @broccoliSpicy in #2626
• perf: skip documents with WAND by @BubbleCal in #2632

Other Changes

• refactor: move planner.rs, logical_expr.rs, and sql.rs from lance to lance_datafusion by @westonpace in #2622
• refactor: remove custom ProjectionExec by @wjones127 in #2608

New Contributors

• @BitPhinix made their first contribution in #2619

Full Changelog: v0.14.1...v0.15.0

v0.15.0-beta.2

What's Changed

New Features 🎉

• feat: add FSST string compression by @broccoliSpicy in #2470
• feat: integrate inverted index into lance index APIs by @BubbleCal in #2577
• feat: add projection to take and take_rows by @westonpace in #2623

Bug Fixes 🐛

• fix: upgrade AWS SDK by @wjones127 in #2616
• fix: specify behavior version when creating dynamodb config by @westonpace in #2624

Other Changes

• refactor: move planner.rs, logical_expr.rs, and sql.rs from lance to lance_datafusion by @westonpace in #2622

Full Changelog: v0.15.0-beta.1...v0.15.0-beta.2

v0.15.0-beta.1

What's Changed

Breaking Changes 🛠

• feat: upgrade DataFusion, Arrow, PyO3, ObjectStore by @wjones127 in #2594

New Features 🎉

• feat: loosen pyarrow pin by @wjones127 in #2597

Performance Improvements 🚀

• perf: improve inverted index performance by @BubbleCal in #2574
• perf: improve v2 scan performance by @westonpace in #2604

Full Changelog: v0.14.2-beta.1...v0.15.0-beta.1

v0.14.2-beta.1

What's Changed

New Features 🎉

• feat: add fast search to to_table method by @chebbyChefNEQ in #2591

Bug Fixes 🐛

• fix: avoid making empty range requests as these will err on S3 by @westonpace in #2592

Full Changelog: v0.14.1...v0.14.2-beta.1

v0.14.1

What's Changed

New Features 🎉

• feat: add inverted index by @BubbleCal in #2526
• feat(python): expose fast search flag in python by @eddyxu in #2573
• feat: use hyperloglog for cardinality estimation for dictionary encoding by @niyue in #2555
• feat: add scalar bitmap index to lance by @raunaks13 in #2560
• feat: add a standalone method for calculating the PQ model by @westonpace in #2562
• feat: integrate bitmap index into top-level lance APIs by @westonpace in #2575
• feat: support to set ef for search from python by @BubbleCal in #2569
• feat: dataset update accepts binary value by @eddyxu in #2579
• feat: do flat search if too many rows are filtered out by @BubbleCal in #2583
• feat: add label list index by @westonpace in #2581
• feat: allow passing in precomputed centroids to lance.util.KMeans by @jiachengdb in #2586
• feat: add standalone partition assignment operation by @westonpace in #2556

Bug Fixes 🐛

• fix: add support for list fields with names other than 'item' by @westonpace in #2580
• fix: propagate storage_options to v2 writer by @jiachengdb in #2578
• fix: be compatible with v1 index format by @BubbleCal in #2582
• fix: search panic with prefilter and HNSW by @BubbleCal in #2584

Performance Improvements 🚀

• perf: change scalar index to return RowIdTreeMap instead of u64 array by @westonpace in #2587

Full Changelog: v0.14.0...v0.14.1

v0.14.1-beta.2

What's Changed

New Features 🎉

• feat: support to set ef for search from python by @BubbleCal in #2569

Full Changelog: v0.14.1-beta.1...v0.14.1-beta.2

v0.14.1-beta.1

What's Changed

New Features 🎉

• feat: add inverted index by @BubbleCal in #2526
• feat(python): expose fast search flag in python by @eddyxu in #2573
• feat: use hyperloglog for cardinality estimation for dictionary encoding by @niyue in #2555
• feat: add scalar bitmap index to lance by @raunaks13 in #2560
• feat: add a standalone method for calculating the PQ model by @westonpace in #2562
• feat: integrate bitmap index into top-level lance APIs by @westonpace in #2575

Full Changelog: v0.14.0...v0.14.1-beta.1

v0.14.0

What's Changed

Breaking Changes 🛠

• fix: return correct sub index statistics by @BubbleCal in #2532
• feat: change to snapshot isolation by @wjones127 in #2537

New Features 🎉

• feat: enhance binary array encoding, make it the default by @westonpace in #2521
• feat: stable row id support in queries by @wjones127 in #2452
• feat: make it possible to create v2 fragments using the fragment API by @westonpace in #2534
• feat: add dictionary encoding to lance by @raunaks13 in #2409
• feat: provide a fast ANN search plan by @eddyxu in #2554
• feat: create standalone IVF training API in python by @westonpace in #2553
• feat: support u8 for existing distance types by @BubbleCal in #2501
• feat: reserve cores for IO by @chebbyChefNEQ in #2496
• feat: allow users to create a file writer without a schema by @westonpace in #2558

Bug Fixes 🐛

• fix: pytorch dot inference pass unwanted parameters by @eddyxu in #2539
• fix: show underlying cause of DynamoDB service error by @wjones127 in #2541
• fix: build HNSW with wrong distance type by @BubbleCal in #2533
• fix: propagate storage_options to LanceFragment.create and LanceDataset.commit by @jiachengdb in #2547
• fix: propagate shuffle_partition_concurrency from Python SDK by @jiachengdb in #2564
• fix: avoid multiple threads loading same index partition by @albertlockett in #2559
• fix: failed to return index stats if the first partition is empty by @BubbleCal in #2565
• fix: propagate storage_options to create_index for reading precomputed partition file by @jiachengdb in #2568

Documentation 📚

• docs: artefact_management_doc by @tanaymeh in #2504

Performance Improvements 🚀

• perf: avoid allocating bitvec for each insertion while indexing HNSW by @BubbleCal in #2527

Other Changes

• refactor: move IVF_HNSW_SQ & IVF_FLAT to new buliding & search path by @BubbleCal in #2469
• refactor: flat search to use datafusion top k by @eddyxu in #2535
• refactor: remove generic PQ impl and trait by @BubbleCal in #2529

New Contributors

• @jiachengdb made their first contribution in #2547
• @walterddr made their first contribution in #2382

Full Changelog: v0.13.0...v0.14.0

v0.14.0-beta.2

What's Changed

New Features 🎉

• feat: add dictionary encoding to lance by @raunaks13 in #2409
• feat: provide a fast ANN search plan by @eddyxu in #2554
• feat: create standalone IVF training API in python by @westonpace in #2553
• feat: support u8 for existing distance types by @BubbleCal in #2501
• feat: reserve cores for IO by @chebbyChefNEQ in #2496
• feat: allow users to create a file writer without a schema by @westonpace in #2558

Bug Fixes 🐛

• fix: propagate shuffle_partition_concurrency from Python SDK by @jiachengdb in #2564
• fix: avoid multiple threads loading same index partition by @albertlockett in #2559

Other Changes

• refactor: remove generic PQ impl and trait by @BubbleCal in #2529

New Contributors

• @walterddr made their first contribution in #2382

Full Changelog: v0.14.0-beta.1...v0.14.0-beta.2

v0.14.0-beta.1

What's Changed

Breaking Changes 🛠

• fix: return correct sub index statistics by @BubbleCal in #2532
• feat: change to snapshot isolation by @wjones127 in #2537

New Features 🎉

• feat: stable row id support in queries by @wjones127 in #2452
• feat: make it possible to create v2 fragments using the fragment API by @westonpace in #2534

Bug Fixes 🐛

• fix: pytorch dot inference pass unwanted parameters by @eddyxu in #2539
• fix: show underlying cause of DynamoDB service error by @wjones127 in #2541
• fix: build HNSW with wrong distance type by @BubbleCal in #2533
• fix: propagate storage_options to LanceFragment.create and LanceDataset.commit by @jiachengdb in #2547

Performance Improvements 🚀

• perf: avoid allocating bitvec for each insertion while indexing HNSW by @BubbleCal in #2527

Other Changes

• refactor: flat search to use datafusion top k by @eddyxu in #2535

New Contributors

• @jiachengdb made their first contribution in #2547

Full Changelog: v0.13.1-beta.1...v0.14.0-beta.1