As you may know, one of my (very geeky) interests is Approximate nearest neigbor methods, and I'm the author of a Python package called Annoy. I've also built a benchmark suite called ann-benchmarks to compare different packages.

Erik Bernhardsson

An update on the approximate nearest neighbor (ANN) landscape, focusing on the ann-benchmarks suite and how Annoy now ranks fourth in speed at high precision levels. Two libraries — FALCONN (LSH-based) and NMSLIB (with SW-graph and hnsw algorithms) — have surpassed Annoy, with hnsw being nearly an order of magnitude faster. The post also highlights a new paper proposing a t-SNE improvement for large high-dimensional dataset visualization that uses Annoy, and mentions an upcoming talk at the DataPhilly meetup.

Approximate nearest news