minimalKNN 0.4

MinimalKNN: minimal package to construct k-NN Graph

A Minimal k-Nearest Neighbor Graph Construction Library

Overview

This package will provide a function to construct an approximated k-Nearest Neighbor graph from a list of three dimensional points. The graph construction algorithm is based on NN-descent presented in Dong, Moses, & Li (2011)[^DML2011]. The Euclidean and Manhattan metrics are implemented in the current version, while only the Euclidean one is available in Python. The algorithm efficiently constructs an approximated k-Nearest Neighbor graph. This provides a portable C++11 header and a Python interface.

Dependencies

The library is written in C++11 and do not depends on any library outside of the STL. The Python interface is depends on NumPy, and functional test procedures depend on Matplotlib. The library is developed on g++ version 5.4 installed in Linux Mint 18.1 (serena). The Python interface is developed on Python 3.7.1 and Numpy 1.18.1.

References

[^DML2011]: Wei Dong, Charikar Moses, & Kai Li, WWW'11: Proceedings of the 20th international conference on World wide web (2011), 577--586 (doi: 10.1145/1963405.1963487)