NMSLIB

NMSLIB (Non-Metric Space Library) is an efficient and extensible similarity search library and toolkit for high-dimensional vector spaces, with a particular focus on generic and non-metric spaces. It is widely used for approximate nearest neighbor (ANN) search and evaluation of k-NN methods.

• Website/Source: https://github.com/nmslib/nmslib
• License: Apache-2.0
• Category: SDKs & Libraries
• Tags: open-source, ANN, similarity-search, high-dimensional

Features

• Efficient Similarity Search: Provides fast search methods in high-dimensional and non-metric spaces.
• Extensible Architecture: Supports adding new search methods and distance functions.
• Multiple Algorithms: Implements various algorithms for approximate nearest neighbor search, including:
  • Hierarchical Navigable Small World Graphs (HNSW)
  • VP-tree (Vantage Point tree)
  • Neighborhood Approximation index (NAPP)
  • Inverted file based methods
  • k-NN graph construction (NN-Descent)
• Metric and Non-Metric Spaces: Principled support for both metric and non-metric space search.
• Cross-Platform: Usable on multiple platforms with no third-party dependencies in the core library.
• Language Support:
  • Native C++ API
  • Python bindings
  • Query server for Java and other languages via Apache Thrift (Java native client available)
• Benchmarking Toolkit: Includes tools to evaluate and benchmark k-NN methods.
• Standalone HNSW: Fastest method (HNSW) also available as a header-only standalone library.
• Open Source: Actively maintained, with contributions from multiple authors and a strong community.
• Used in Production: Integrated into services like Amazon Elasticsearch Service.

Pricing

NMSLIB is open-source and free to use under the Apache-2.0 license.