Module hnsw 

HNSW (Hierarchical Navigable Small World) Index

Production-quality implementation of the HNSW algorithm for approximate nearest neighbor search in high-dimensional vector spaces.

Algorithm Overview

HNSW builds a multi-layer graph structure where:

• Layer 0 contains all vectors
• Higher layers contain progressively fewer vectors (exponentially decaying)
• Each layer is a navigable small world graph with bounded degree
• Search proceeds from top layer down, greedy navigating to nearest neighbors

Performance Characteristics

• Search: O(log n) approximate nearest neighbor queries
• Insert: O(log n) amortized insertion time
• Space: O(n * M) where M is max connections per layer
• Accuracy: Configurable via ef_construction and ef_search parameters

References

• Malkov, Y. A., & Yashunin, D. A. (2018). “Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs” IEEE Transactions on Pattern Analysis and Machine Intelligence.

Structs

HnswConfig

HNSW index configuration

HnswIndex

HNSW index for approximate nearest neighbor search

HnswSearchResult

Search result with distance and metadata

HnswStats

Statistics about the HNSW index structure

Enums

DistanceMetric

Distance metrics supported by HNSW