weka的java使用——聚类

weka是著名的数据挖掘工具，在这里有详细介绍，IDMer老师的博客里也有比较详细的用法描述。当然，如果直接使用weka的工具，自然没有问题，但是如果想用weka的功能在自己的平台框架中呢？我这里放出一个当初对weka的源码学习过程，主要是如何调用weka的api。仅供参考，代码中有什么问题，欢迎邮件联系。

这里简单讲解一下流程。构造方法首先载入一个arff文件，然后调用doCluster（）方法进行聚类。本文用到的arff文件是weka的data目录里自带的标准数据集。主要格式如下：

1/**
  2 * 
  3 */
  4package edu.tju.ikse.mi.util;
  5
  6import java.io.File;
  7import java.io.FileNotFoundException;
  8import java.io.IOException;
  9import java.io.PrintWriter;
 10import java.util.Scanner;
 11
 12import edu.tju.ikse.mi.anno.util.CfUtil;
 13
 14
 15import weka.clusterers.XMeans;
 16import weka.core.Instances;
 17import weka.core.converters.ArffLoader;
 18
 19/**
 20 * @author Jia Yu
 21 * @date 2010-5-28
 22 */
 23public class WekaCluster {
 24
 25    /**
 26     * @param args
 27     */
 28
 29    private ArffLoader loader;
 30    private Instances dataSet;
 31    private weka.clusterers.Clusterer cluster;
 32    private int numOfClusters;
 33    private String newAttribute;
 34    private File arffFile;
 35    private int sizeOfDataset;
 36
 37    public WekaCluster(File arffFile) {
 38        this.arffFile = arffFile;
 39        doCluster();
 40    }
 41
 42    private void doCluster() {
 43        loader = new ArffLoader();
 44        newAttribute = "";
 45        try {
 46            loader.setFile(arffFile);
 47            dataSet = loader.getDataSet();
 48            cluster = new XMeans();
 49            cluster.buildClusterer(dataSet);
 50            numOfClusters = cluster.numberOfClusters();
 51            StringBuilder sb = new StringBuilder();
 52            for (int i = 0; i < numOfClusters; i++) {
 53                sb.append("s" + (i + 1) + " ");
 54            }
 55            newAttribute = sb.toString().trim();
 56            sizeOfDataset = dataSet.numInstances();
 57        } catch (Exception e) {
 58            e.printStackTrace();
 59        }
 60    }
 61
 62    public void newArffWriter() {
 63        int lineNum = 0;
 64        try {
 65            Scanner input = new Scanner(arffFile);
 66            PrintWriter out = new PrintWriter(CfUtil
 67                    .GetFileNameNoExtFromFileName(arffFile.getName())
 68                    + "_classification.arff");
 69
 70            while (input.hasNext()) {
 71                String line = input.nextLine();
 72                if (line.startsWith("@relation")) {
 73                    out.println("@relation" + line.substring(9)
 74                            + "_classification");
 75                } else if (line.startsWith("@data")) {
 76                    out.println("@attribute shape {" + newAttribute + "}");
 77                    out.println("@data");
 78                } else if (line.startsWith("@attribute")) {
 79                    out.println(line);
 80                } else if (line.isEmpty()) {
 81                    out.println();
 82                } else {
 83                    line += ",class"
 84                            + (cluster.clusterInstance(dataSet
 85                                    .instance(lineNum)) + 1);
 86                    out.println(line);
 87                    lineNum++;
 88                }
 89            }
 90            out.close();
 91        } catch (FileNotFoundException e) {
 92            e.printStackTrace();
 93        } catch (Exception e) {
 94            e.printStackTrace();
 95        }
 96    }
 97
 98    public int clusterNewInstance(weka.core.Instance instance) {
 99        int indexOfCluster = -1;
100        try {
101            indexOfCluster = cluster.clusterInstance(instance);
102            //System.out.println("cluster " + indexOfCluster);
103        } catch (Exception e) {
104            e.printStackTrace();
105        }
106        return indexOfCluster;
107    }
108
109    public double[] frequencyOfCluster() {
110        int[] sum = new int[this.numOfClusters];
111        try {
112            for (int i = 0; i < this.sizeOfDataset; i++) {
113                sum[cluster.clusterInstance(dataSet.instance(i))]++;
114            }
115        } catch (Exception e) {
116            e.printStackTrace();
117        }
118        double[] fre = new double[sum.length];
119        for (int i = 0; i < sum.length; i++) {
120            fre[i] = (double)sum[i] / (double)this.sizeOfDataset;
121        }
122        return fre;
123    }
124
125    public static void main(String[] args) {
126        File file = new File("cpu.arff");
127        WekaCluster wc = new WekaCluster(file);
128        double[] fre = wc.frequencyOfCluster();
129        for(int i=0;i<fre.length;i++)
130            System.out.println(fre[i]);
131        // wc.newArffWriter(file);
132        double[] feature = { 125,256,6000,256,16,128,199 };
133        weka.core.Instance ins = new weka.core.Instance(7);
134        for (int i = 0; i < ins.numAttributes(); i++) {
135            ins.setValue(i, feature[i]);
136            // System.out.println(ins.attribute(i).getLowerNumericBound());
137        }
138        System.out.println("cluster in : "+wc.clusterNewInstance(ins));
139    }
140
141}
142

@relation ‘cpu’
@attribute MYCT real
@attribute MMIN real
@attribute MMAX real
@attribute CACH real
@attribute CHMIN real
@attribute CHMAX real
@attribute class real
@data
125,256,6000,256,16,128,199
29,8000,32000,32,8,32,253
29,8000,32000,32,8,32,253
这里摘取了3项。运行程序执行结果如下：
0.03827751196172249
0.16267942583732056
0.69377990430622
0.10526315789473684
cluster in : 0
表示聚类方法将数据集聚为四类，程序中提供的instance被聚到第一类里。每一类的在总文件中的比率如上显示。
具体的数据挖掘的内容就不在这里讲述了。只是为大家提供一个weka的java用法实现。方便在程序中使用weka。