shackenberg/cukmeans.py

Implementation of the kmeans algorithm using pycuda

cukmeans.py - Kmeans in PyCUDA

This is an implemenations of the [clustering algorithm k-means][2] in [PyCUDA][3]. As the interface was taken from [Scipy K-means][1], one can easily replace all usages.

Sample usage

from cukmeans import cukmeans
from numpy.random import randn
nclusters = 100
npoints = 10000
dimensions = 2
data = randn(npoints, dimensions)
gpu_book, gpu_dist = cukmeans(data, nclusters)

Demo and benchmark modus

If you run the script directly it starts into a demo modus benchmarking itself against the C based Scipy implementation of kmeans and a kmeans implementation using flann (if available).

Contents

• cukmeans.py the kmeans implementation
• flkmeans.py a flann based kmeans implemantation as additional benchmark

Requirements

• PyCUDA: http://mathema.tician.de/software/pycuda
• Flann (optional*): https://github.com/mariusmuja/flann

* Flann is only needed as an aditional kmeans implementation to benchmark to. [1]: https://github.com/scipy/scipy/blob/master/scipy/cluster/vq.py [2]: https://en.wikipedia.org/wiki/K-means_clustering [3]: http://mathema.tician.de/software/pycuda