Repository navigation
Vector API: find_nearest filters and groups, build options, sparse index helper #382
Description
The engine's vector index has options our Python API does not expose, all stable upstream: in vector.neighbors the where, groupBy, groupSize and maxDistance options (where is new in 26.11.1, so it needs a version gate), and in LSM_VECTOR the neighborOverflowFactor and alphaDiversityRelaxation build settings. The sparse vector index (LSM_SPARSE_VECTOR, vector.sparseNeighbors) is reachable only through SQL, and IndexType lacks LSM_SPARSE_VECTOR. Our create_vector_index defaults add_hierarchy to true while the engine defaults it to false; the docstring should say so (the benchmark records the value it uses). Plan: add these to find_nearest, create_vector_index and a small sparse helper, with tests, examples and docs in the same change.
Activity
tae898 commentedon Oct 21, 2026
Scope narrowed to match the bindings' DSL-first design (
docs/api-access-methods.md): no new wrapper API. The vector options (where,groupBy,groupSize,maxDistance,neighborOverflowFactor,alphaDiversityRelaxation) and the sparse index are documented as SQL (vector.neighbors,vector.sparseNeighbors,CREATE INDEX ... METADATA) with tests where a documented form is not covered yet, plus theadd_hierarchydefault note andIndexType.LSM_SPARSE_VECTORif the enum is part of existing calls.tests: the where option asserts on both engines instead of skipping; …
Document the vector SQL options and add IndexType.LSM_SPARSE_VECTOR (#…