weka.clusterers

Class ClusterUtils

java.lang.Object

weka.clusterers.ClusterUtils

public class ClusterUtils
extends java.lang.Object

Some static utils methods for clustering

Version:

$Revision: $

Author:

Mark Hall (mhall{[at]}pentaho{[dot]}com)

Method Summary

Modifier and Type	Method and Description
static weka.core.Instances	getPrimingDataForDistanceFunction(weka.core.Instances headerWithSummary) Compute priming data for a distance function (i.e.
static java.util.List<weka.core.Instances>	weightSketchesAndClusterToFinalStartPoints(int numRuns, int numClusters, CentroidSketch[] finalSketches, KMeansReduceTask[] statsForSketches, boolean debug) Utility method to perform the last phase of the k-means|| initialization - i.e.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

weightSketchesAndClusterToFinalStartPoints

public static java.util.List<weka.core.Instances> weightSketchesAndClusterToFinalStartPoints(int numRuns,
                                                                                             int numClusters,
                                                                                             CentroidSketch[] finalSketches,
                                                                                             KMeansReduceTask[] statsForSketches,
                                                                                             boolean debug)
                                                                                      throws DistributedWekaException

Utility method to perform the last phase of the k-means|| initialization - i.e. takes the final centroid sketches and the clustering stats that show how many training points are assigned to each instance (candidate center) in each sketch, and then weights the sketch centroids and runs a local KMeans on the weighted data to get the final requested number of starting points.

Parameters:

numRuns - number of runs of k-means being done

numClusters - the requested number of clusters (and thus starting points for k-means)

finalSketches - the final centroid sketches produced by the first phase of k-means||

statsForSketches - the clustering statistics for the final sketches

debug - true if debugging info is to be output

Returns:

a final list of starting points (one set of instances for each run of k-means to be performed)

Throws:

DistributedWekaException - if a problem occurs

getPrimingDataForDistanceFunction

public static weka.core.Instances getPrimingDataForDistanceFunction(weka.core.Instances headerWithSummary)
                                                             throws DistributedWekaException

Compute priming data for a distance function (i.e. data that can be used to intialize the mins and maxes of numeric variables for normalization purposes).

Parameters:

headerWithSummary - header with summary attributes

Returns:

a two-instance data set containing min and maxes for all numeric attributes. Nominal attributes will have missing values

Throws:

DistributedWekaException - if a problem occurs