Name: h2oai-kmeans
Owner: H2O.ai
Description: kmeans clustering with multi-GPU capabilities
Forked from: NVIDIA/kmeans
Created: 2017-06-07 00:13:15.0
Updated: 2017-06-08 00:25:25.0
Pushed: 2017-06-22 17:27:40.0
Homepage: null
Size: 50
Language: C++
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
A simple kmeans clustering implementation for double precision data, written for CUDA GPUs.
There are two ideas here:
The CUDA code here is purposefully non-optimized - this code is not meant to be the fastest possible kmeans implementation, but rather to show how using libraries like thrust and BLAS can provide reasonable performance with high programmer productivity.
This version has been updated to use multiple GPUs attached to the same machine. You do not need to specify the number of GPUs, the program will detect and use them.
To build, edit Makefile to specify CUB_HOME, the location of your CUB files Then call make.
A simple test case is run when you invoke the executable 'test'.
For demonstration, test will generate and solve 3 test cases of different sizes. At the prompt, specify 't' for a tiny test case, 'm' for a slightly bigger test case, and 'h' for a huge test case: 1 million points, with 50 dimensions and 100 clusters, for 50 iterations.