pyTree : a collection of python codes for tree-based neighbor search

author : Jake Vanderplas vanderplas@astro.washington.edu

This is mainly a test-bed for codes to be included in scikit-learn. As such, it's not set up to be a fully-fledged package, but to be installed in-place.

Compiling

This is set-up for in-place compilation. Compile with the command

python setup.py build_ext --inplace

Contents

cpp_balltree : This is the C++ BallTree code which I submitted to scikit-learn. The cython wrapper and a few tweaks to the C++ source are due to some of the scikit-learn contributers

npy_balltree : This is an attempt to speed-up and streamline the ball tree implementation Its approach is to use no dynamic memory allocation, but instead store all the info for the built ball tree in a collection of numpy arrays. I'm also making the ball tree more flexible (any minkowski p-distance can be used). This is a work in progress. Currently initializing the tree is approximately 8-10 times faster than in the C++ version. The query() function is about 50% faster. The query_radius() function has not yet been implemented.