v0.1.0 - Initial Release

superbit_lsh v0.1.0

A lightweight, in-memory vector index for approximate nearest-neighbor (ANN) search using Locality-Sensitive Hashing.

Highlights

• Random hyperplane LSH (SimHash) for cosine, Euclidean, and dot-product similarity
• Multi-probe querying for improved recall without extra tables
• Thread-safe concurrent access via parking_lot::RwLock
• Builder pattern for ergonomic index configuration
• Auto-tuning -- suggest_params recommends optimal parameters for a target recall
• Runtime metrics -- lock-free atomic counters for query latency, candidates, and hit rates

Optional Features

Flag	Effect
parallel	Parallel bulk insert and batch query via rayon
persistence	Save/load index to disk (serde + bincode + JSON)
python	Python bindings via PyO3
full	Enables parallel + persistence

Install

[dependencies]
superbit_lsh = "0.1"

Performance

Benchmarked on 768-dimensional vectors (release mode):

Dataset Size	LSH Query	Brute-force	Speedup
10k vectors	4.6 us	1.04 ms	226x
100k vectors	34.5 us	60.6 ms	1,756x

Links

• crates.io
• docs.rs