| name | 0.2.3-understand-idx_hybrid_composition |
| description | [0.2.3] Strategy combining multiple index types such as HNSW over PQ or IVF with scalar quantization to balance speed |
understand-idx_hybrid_composition
CALL NUMBER: deep_nearest_neighbor_sea.idx_hybrid_composition : vector_databases_and_embeddings(5)
DEFINITION: Strategy combining multiple index types such as HNSW over PQ or IVF with scalar quantization to balance speed accuracy and memory
Invoke this skill to understand idx_hybrid_composition down to its primitives. The RELATIVE ROOT below is the least-fixed-point closure of everything it bundles from — the full import cone, grouped by the lib each prim comes from. Projected from a prover-typed KB (MAP/SWI-Prolog consistency gate): every reference below resolves.
THE RELATIVE ROOT (the import cone, by lib)
from vector_databases_and_embeddings
- hnsw_index (d1): Hierarchical Navigable Small World graph; layered skip-list + NN graph enabling logarithmic-layer traversal then constant-time local search.
- ivf_index (d1): Inverted File Index; clusters vectors into Voronoi cells via k-means; search prunes to nearest centroids then scans partitions.
- product_quantization (d1): A vector quantization technique that splits a high-dimensional vector into subvectors, independently quantizes each subvector, and concatenates codes to form a compact representation.
- indexing_strategy (d2): Choice and configuration of ANN algorithm (HNSW, IVF, ANNOY, LSH) and its parameters (ef, nlist, nprobe) governing recall/speed trade-off.
- memory_footprint (d2): RAM or VRAM required to hold index and vectors; quantization (PQ, int8) is the primary compression lever.
Projected from the vector databases and embeddings KB (262 concepts / 150 relations) — consistency-typed by MAP; the facet list after the colon IS the cross-lib dependency web.
(leaf — this is an actual skill.)
