PANNA: Playground for Approximate Nearest Neighbor Algorithms

This repository aims at providing useful building blocks to implement algorithms for approximate nearest neighbor search.

The repository provides mainly two things:

• A software library to build similarity search algorithms, implemented mainly as a C++ library, with python bindings
• An environment to run experiments, pinning versions and dependencies of related baselines by means of Nix flakes

Python: installing

The library should be easily installable with the usual

pip install pypanna

If you are using uv to manage your dependencies

uv add pypanna

should suffice.

C++: Building

This is, first and foremost, a header only library requiring C++20 and depending on ffht (vendored in external). To integrate with other codebases simply place include/panna in your include path, while making sure that the headers of the dependency (i.e. the contents of external) are included as well.

That said, the repository includes tests and examples, which are built using cmake with the usual steps

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make