An implementation of HNSW (Hierarchical Navigable Small World) algorithm for approximate nearest neighbor search. This is not directly based on the original paper, it is simplified and easy to understand, while still being reasonably efficient and robust, I guess

Checkout this repo: blaze-db, which is a vector database built on top of this HNSW implementation.

How to bench?

cargo run --example bench --release -- ../../datasets/dim1536_size1M 4
                                         <path to dataset> <num of files to read>

Make sure to have cargo & this dataset

Ref:

• https://en.wikipedia.org/wiki/Curse_of_dimensionality
• https://arxiv.org/pdf/1603.09320
• https://arxiv.org/abs/2512.06636
• https://arxiv.org/pdf/2406.03482
• https://arxiv.org/html/2412.01940v1
• https://www.pinecone.io/learn/series/faiss/hnsw/
• https://www.techrxiv.org/users/922842/articles/1311476-a-comparative-study-of-hnsw-implementations-for-scalable-approximate-nearest-neighbor-search

Note: Some of the ref are of my TODO list, I have not read them yet, but I think they are relevant, so I put them here for future reference.