0.10.3

What's Changed

• Disable consolidate_update_fragments for TileDB Cloud URIs by @NikolaosPapailiou in #536
• Improve ObjectAPI documentation by @NikolaosPapailiou in #538
• Decouple consolidate and vacuum operations by @NikolaosPapailiou in #539
• Parallelize asset creation and registration by @NikolaosPapailiou in #540
• Add query with driver mode for ObjectIndex by @NikolaosPapailiou in #541
• [automated] Update backwards-compatibility-data for release 0.9.0 by @github-actions in #527
• [automated] Update backwards-compatibility-data for release 0.10.0 by @github-actions in #530
• Improve timers by @jparismorgan in #496
• Add more instructions on building C++ by @jparismorgan in #523
• Support multiple embeddings per object in Object API by @NikolaosPapailiou in #542
• Add ColPali embedding and support for processing PDFs as images by @NikolaosPapailiou in #543
• Add a new read_vector() helper which does a multi-range query by @jparismorgan in #545
• Rename group_uri to index_uri by @jparismorgan in #544
• Refactor group version check by @jparismorgan in #547
• Cleanup unit_ivf_pq_index.cc by @jparismorgan in #548
• Update tdbMatrixMultiRange to support both slices and individual indices by @jparismorgan in #550
• Add start, stop, and temporal_policy to FeatureVector and FeatureVectorArray by @jparismorgan in #549
• Rename nlist to partitions by @jparismorgan in #551
• Fix types in qv_partition_with_scores() and train_no_init() by @jparismorgan in #546
• Various small code cleanups from IVF_PQ OOC work by @jparismorgan in #552
• Fix AVX2 distance function and build with AVX2 if available on host machine by @jparismorgan in #524
• Add numpy 2 support by @jparismorgan in #434
• Add bioimage search utils and example notebook by @NikolaosPapailiou in #553
• Support out-of-core and distributed IVF_PQ ingestion by @jparismorgan in #531
• Update IVF_PQ to use temp directory pattern to support parallel ingestion by @jparismorgan in #554
• Write dimensions as uint64 in Python by @jparismorgan in #556
• Add IVF_PQ backwards compatibility data generation and testing by @jparismorgan in #555
• Remove warning that IVF_PQ index is not supported by @jparismorgan in #557

Full Changelog: 0.10.2...0.10.3

0.10.2

What's Changed

• Update to TileDB Core 2.26.2 by @jparismorgan in #535

Full Changelog: 0.10.1...0.10.2

0.10.1

What's Changed

• Add explicit markdown parser for DirectoryTextReader by @NikolaosPapailiou in #532

Full Changelog: 0.10.0...0.10.1

0.10.0

What's Changed

• Update TileDB Core 2.26.1 by @jparismorgan in #528

Full Changelog: 0.9.0...0.10.0

0.9.0

What's Changed

• Fix bug for Index clear_history by @NikolaosPapailiou in #522
• Add open method for Index class by @NikolaosPapailiou in #503
• Fix small C++ warnings around type conversions by @jparismorgan in #520
• Update IVF_PQ to set memory_budget in constructor, support preload feature_vectors and metadata only modes by @jparismorgan in #518
• Update TileDB Core 2.26.0 by @NikolaosPapailiou in #521

Full Changelog: 0.8.1...0.9.0

0.8.1

What's Changed

• Update local-benchmarks script to save results to a new directory during each run by @jparismorgan in #464
• [automated] Update backwards-compatibility-data for release 0.8.0 by @github-actions in #467
• Fix C++ warnings across codebase by @jparismorgan in #461
• Enable more benchmarks in ann-benchmarks script by @jparismorgan in #470
• Optimize distance computations for IVFPQ by @NikolaosPapailiou in #468
• Fix C++ incorrect or ambiguous primitive types by @jparismorgan in #455
• Add C++ sanitizers by @jparismorgan in #471
• Use single shared seed for rng generation in C++ by @jparismorgan in #469
• Make logging utils thread-safe by @jparismorgan in #474
• Add nightly CI which runs sanitizers by @jparismorgan in #476
• Add new test and cleanup C++ code by @jparismorgan in #475
• Fix counting_sum_of_squares_distance thread sanitizer error by @jparismorgan in #472
• Parallelize Vamana query by @jparismorgan in #463
• Fix address sanitizer error when storing metadata in destructor by @jparismorgan in #473
• Fix mutex lock error in logging utils by @jparismorgan in #478
• Add option to ann-benchmarks.py to skip benchmarks and leave instance running by @jparismorgan in #477
• Add dump() methods to logging classes by @jparismorgan in #479
• Add more testing for Vamana helpers, add option to skip top_k to improve ingestion speed by @jparismorgan in #481
• Avoid copies in query code by @jparismorgan in #483
• Add Vamana storage spec by @jparismorgan in #484
• Add more comments about IVF parameters partitions and nprobe by @jparismorgan in #487
• IVF_FLAT Distance metric by @cainamisir in #451
• Update matrix to take in a std::vector instead of std::initializer_list by @jparismorgan in #492
• Fix flaky IVF_PQ Python test and add more C++ testing around pq-encoding by @jparismorgan in #494
• Fix infinite loop in kmeans by @jparismorgan in #491
• Distance metric integration for vamana, and refactoring of distance metrics by @cainamisir in #460
• Add out-of-core query() support to IVF PQ by @jparismorgan in #485
• IVF_PQ: Remove unused pq_ivf_centroids, remove extra call to train_ivf(), add comments, cleanup code by @jparismorgan in #489
• Refactor logging helpers by @jparismorgan in #497
• Name each benchmark within local-benchmarks.py by @jparismorgan in #498
• Add scGPT and scvi embeddings and improve SOMA reader by @NikolaosPapailiou in #501
• Vlad/l2 sumofsquares by @cainamisir in #486
• Update ann-benchmarks.py to connect to running instance by @jparismorgan in #500
• Fix some TODOs after the cloud release by @NikolaosPapailiou in #505
• Various small tdb matrix cleanups by @jparismorgan in #510
• Distance metric small fixes: uninitialized value in IVF_PQ and pass setting during consolidate_updates() by @jparismorgan in #508
• Add fixed_min_triplet_heap by @jparismorgan in #507
• Update IVF_PQ array names by @jparismorgan in #511
• Check finite and infinite IVF_PQ queries return the same ids and distances, fix count_intersections() to not modify inputs, update read_index_finite() to return data by @jparismorgan in #509
• Add new plot to local-benchmarks.py showing results from all indexes by @jparismorgan in #512
• DirectoryReader: set text/plain as default mime type if it is not found by @NikolaosPapailiou in #513
• Add option to run local-benchmarks.py with index at a tiledb uri by @jparismorgan in #514
• IVF_PQ re-ranking by @jparismorgan in #502
• Support AWS index URI in local-benchmarks.py by @jparismorgan in #516
• Add k_factor to local-benchmarks.py by @jparismorgan in #517

Full Changelog: 0.8.0...0.8.1

0.8.0

What's Changed

• Remove use of set_coords_filter_list from dense array creation by @jparismorgan in #439
• [automated] Update backwards-compatibility-data for release 0.7.0 by @github-actions in #438
• Tune default ingestion configuration to avoid OOM errors by @NikolaosPapailiou in #440
• Add ids to Python FeatureVectorArray by @jparismorgan in #442
• Add Optional to Python code that was missing it by @jparismorgan in #443
• For type-erased Python indexes, 1) Don't consolidate parts and ids Arrays 2) Avoid extra Schema open in constructor by @jparismorgan in #444
• Cleanups to ivf_index() C++ code, and small cleanups in Python by @jparismorgan in #430
• Support IVF PQ consolidation by storing raw feature vectors and external IDs by @jparismorgan in #447
• tdbPartitionedMatrix will automatically close Array's when done reading by @jparismorgan in #448
• Re-enable IVF PQ tests by @jparismorgan in #450
• Save kmeans settings to IVF PQ metadata by @jparismorgan in #452
• Allow setting IVF PQ partitions when re-ingesting, fix IVF PQ object index tests by @jparismorgan in #453
• Avoid creating one temp array for each ingestion work item by @NikolaosPapailiou in #449
• Add Vector Search storage format spec by @NikolaosPapailiou in #456
• Fix markdown format for storage spec by @NikolaosPapailiou in #457
• Update dimensions to be uint64_t in C++ by @jparismorgan in #454
• Configure memory budget for distributed OOC queries by @NikolaosPapailiou in #462
• Add local benchmarking script by @jparismorgan in #459
• Distance metrics integration by @cainamisir in #422
• Close tdbMatrix and tdbMatrixWithIds Array's when we have nothing left to to read by @jparismorgan in #466
• Update to TileDB Core 2.25.0 by @jparismorgan in #465

Full Changelog: 0.7.0...0.8.0

0.7.0

What's Changed

• [automated] Update backwards-compatibility-data for release 0.6.0 by @github-actions in #432
• Remove apis/python/requirements-py.txt by @jparismorgan in #433
• Fix bug where we did not set compression filters when creating TileDB Array's in C++ by @jparismorgan in #436
• Update to TileDB Core 2.24.2 by @jparismorgan in #437

Full Changelog: 0.6.0...0.7.0

0.6.0

What's Changed

• Pin numpy to fix Python CI failures by @NikolaosPapailiou in #419
• Enable OOC processing for IVF_FLAT distributed query execution by @NikolaosPapailiou in #418
• Cleanup IVF PQ C++ index code by @jparismorgan in #421
• Add debug info and remove stripping by @dudoslav in #424
• Fix type-erased indexes writing fragments at timestamp=0, thus fixing IVF PQ time travel by @jparismorgan in #425
• Improve benchmark script - add other vector search libraries and download full results by @jparismorgan in #415
• Remove b_backtrack from Vamana index by @jparismorgan in #428
• Expose Vamana graph building params by @jparismorgan in #423
• Update to TileDB Core 2.24.1 by @jparismorgan in #431

Full Changelog: 0.5.1...0.6.0

0.5.1

What's Changed

• Add script to create an EC2 instance and run ann-benchmarks by @jparismorgan in #412
• [automated] Update backwards-compatibility-data for release 0.5.0 by @github-actions in #414
• Replace asserts with exceptions in C++ code by @jparismorgan in #410
• Add debug symbols in release build by @NikolaosPapailiou in #416
• Fix crash in tdb_matrix_with_ids when there was an empty partition by @jparismorgan in #389

Full Changelog: 0.5.0...0.5.1