N2 is a lightweight, C++ based approximate nearest neighbor (aNN) library designed for high-speed performance on large datasets. It targets researchers and engineers working with vector similarity search, offering an alternative to existing libraries like Annoy and NMSLIB by aiming to combine their strengths and mitigate weaknesses.

How It Works

N2 implements a Hierarchical Navigable Small World (HNSW) graph-based algorithm, a well-established approach for efficient aNN search. Its C++ core is optimized for speed and memory usage, with features like multi-core parallelism for index building and mmap support for handling large index files efficiently. The library supports "angular", "L2" (squared Euclidean), and "dot" product distance metrics.

Quick Start & Requirements

• Install via pip: pip install n2
• Requires Python.
• Official documentation: https://n2.readthedocs.io/en/latest/

Highlighted Details

• Optimized for speed in index build time, search performance, and memory usage.
• Supports multi-core CPUs for index building.
• Integrates mmap by default for efficient handling of large index files.
• Provides Python and Go bindings.

Maintenance & Community

• Developed by Kakao Corp.
• Build status indicated by Travis CI.
• Documentation hosted on Read the Docs.

Licensing & Compatibility

• Licensed under the Apache 2.0 license.
• Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The README notes that N2 version 0.1.6 was used in ann-benchmarks.com as of October 2020, suggesting potential for performance improvements and changes since that benchmark.