/hama-kmeans

KMeans implementation using Hama

Primary LanguageJava

Hama K-Means Clustering
-------------------------------------------------------------------

This project implements k-means clustering using the Hama framework.

-------------------------------------------------------------------
Building

This is a Java Maven project. All dependencies and build configurations are 
located in the ./pom.xml file. Please consult your local environment / IDE on
how to build a Maven project.

The result of a successful build is hama-kmeans-<VERSION>.jar in the ./target
directory.

To run the provided Python display script, you will need to install Python 2.6 
and the matplotlib module.

-------------------------------------------------------------------
Running

You must have a Hama and Hadoop system installation. 
Consult http://wiki.apache.org/hama/GettingStarted.

To run:
$HAMA_HOME/bin/hama jar hama-kmeans.jar ee.ut.cs.willmore.KMeansCluster

The program will generate a random input set on the configured 
HDFS and launch the k-means solver. Output will be stored on HDFS and
additionally be copied to a the local /tmp directory. The location of 
these files will be printed to stdout.

The program support a variety of options for generating input data:

 -points  Number of points (observations). Default value is 1000.
 
 -k       Number of clusters. Default value is the number of BSPPeers.
          Value must not exceed number of BSPPeers.
 
 -display A display script that will be called immediately after clustering has
          completed. A sample Python script is supplied at ./python/graph_output.py
 
 -noise   Number of extra random points scattered throughout the problem space.
 
 -pattern Format of <pattern>[:<size>] where pattern of sphere|cube|random and 
          optional size of points for each cluster. Example is sphere:20 which 
          makes spheres of radius 20.
 
 -size    Point world size in X,Y,Z dimensions. Example: a value of 100 would 
          create a world of 100 X 100 X 100.