In an effort to understand it, I put together a simple, pure python implementation of HNSW, an approximate nearest neighbor library. Learned a lot, and I think for anyone interested in vector search it's an exercise that's absolutely worth doing.
The code is optimized (imo) for readability, and working (albeit, quite slowly) on putting together a tutorial that walks through the motivation and implementation of HNSW.
There's also working code examples for using the library for text and image search with sentence transformers and CLIP!
0 comments