Approximate nearest neighbor search is very useful when you have a large dataset of millions of datapoints and you learn some kind of vector representation of these items. I built a benchmark suite called ann-benchmarks. It pits a number of algorithms in a brutal showdown.

Erik Bernhardsson

Updated benchmark results for approximate nearest neighbor (ANN) search libraries are published, comparing algorithms across multiple datasets (glove-100-angular, sift-128-euclidean, nytimes-256-angular, gist-960-euclidean). New additions include NGT-Panng, pynndescent, and MRPT. The HNSW family of algorithms (via nmslib, hnswlib, and FAISS) dominates the rankings, all developed by Yury Malkov. Benchmarks now enforce single-CPU queries for fairer comparison. Results are also available on an interactive website.

New approximate nearest neighbor benchmarks