kmeans算法Java实现.doc

给点一组 c 点资料 X = {x1, ..., xc} ，每一点都有 d 维；给定一个群聚的数目 k, 求其最好的聚类结果。
（1）package cluster.kmeans;
class Centroid {
    private double mCx, mCy;
    private Cluster mCluster;
    public Centroid(double cx, double cy) {
        this.mCx = cx;
        this.mCy = cy;
    }
    public void calcCentroid() { //only called by CAInstance
        int numDP = mCluster.getNumDataPoints();
        double tempX = 0, tempY = 0;
        int i;
        //caluclating the new Centroid
        for (i = 0; i < numDP; i++) {
            tempX = tempX + mCluster.getDataPoint(i).getX();
            //total for x
            tempY = tempY + mCluster.getDataPoint(i).getY();
            //total for y
        }
        this.mCx = tempX / numDP;
        this.mCy = tempY / numDP;
        //calculating the new Euclidean Distance for each Data Point
        tempX = 0;
        tempY = 0;
        for (i = 0; i < numDP; i++) {
            mCluster.getDataPoint(i).calcEuclideanDistance();
        }
        //calculate the new Sum of Squares for the Cluster
        mCluster.calcSumOfSquares();
    }
    public void setCluster(Cluster c) {
        this.mCluster = c;
    }
    public double getCx() {
        return mCx;
    }
    public double getCy() {
        return mCy;
    }
    public Cluster getCluster() {
        return mCluster;
    }
2）import java.util.Vector;
 class Cluster {
    private String mName;
    private Centroid mCentroid;
    private double mSumSqr;
    private Vector mDataPoints;
    public Cluster(String name) {
        this.mName = name;
        this.mCentroid = null; //will be set by calling setCentroid()
        mDataPoints = new Vector();
    }
    public void setCentroid(Centroid c) {
        mCentroid = c;
    }
    public Centroid getCentroid() {
        return mCentroid;
    }
    public void addDataPoint(DataPoint dp) { //called from CAInstance
        dp.setCluster(this); //initiates a inner call to calc EuclideanDistance() in DP.
        this.mDataPoints.addElement(dp);
        calcSumOfSquares();
    }
    public void removeDataPoint(DataPoint dp) {
        this.mDataPoints.removeElement(dp);
        calcSumOfSquares();
    }
    public int getNumDataPoints() {
        return this.mDataPoints.size();
    }
    public DataPoint getDataPoint(int pos) {
        return (DataPoint) this.mDataPoints.elementAt(pos);
    }
    public void calcSumOfSquares() { //called from Centroid
        int size = this.mDataPoints.size();
        double temp = 0;
        for (int i = 0; i < size; i++) {
            temp = temp + ((DataPoint)
this.mDataPoints.elementAt(i)).getCurrentEuDt();
        }
        this.mSumSqr = temp;
    }
    public double getSumSqr() {
        return this.mSumSqr;
    }
    public String getName() {
        return this.mName;
    }
    public Vector getDataPoints() {
        return this.mDataPoints;
    }
}
（3）
package cluster.kmeans;

public class DataPoint {
    private double mX,mY;
    private String mObjName;
    private Cluster mCluster;
    private double mEuDt;
    public DataPoint(double x, double y, String name) {
        this.mX = x;
        this.mY = y;
        this.mObjName = name;
        this.mCluster = null;
    }
    public void setCluster(Cluster cluster) {
        this.mCluster = cluster;
        calcEuclideanDistance();
    }
    public void calcEuclideanDistance() {
    //called when DP is added to a cluster or when a Centroid is recalculated.
        mEuDt = Math.sqrt(Math.pow((mX - mCluster.getCentroid().getCx()),
                                   2) + Math.pow((mY - mCluster.getCentroid().getCy()), 2));
    }
    public double testEuclideanDistance(Centroid c) {
        return Math.sqrt(Math.pow((mX - c.getCx()), 2) + Math.pow((mY - c.getCy()), 2));
    }
    public double getX() {
        return mX;
    }
    public double getY() {
        return mY;
    }
    public Cluster getCluster() {
        return mCluster;
    }
    public double getCurrentEuDt() {
        return mEuDt;
    }
    public String getObjName() {
        return mObjName;
    }
}
（4）import java.util.Vector;

public class JCA {
    private Cluster[] clusters;
    private int miter;
    private Vector mDataPoints = new Vector();
    private double mSWCSS;
    public JCA(int k, int iter, Vector dataPoints) {
        clusters = new Cluster[k];
        for (int i = 0; i < k; i++) {
            clusters[i] = new Cluster("Cluster" + i);
        }
        this.miter = iter;
        this.mDataPoints = dataPoints;
    }
    private void calcSWCSS() {
        double temp = 0;
        for (int i = 0; i < clusters.length; i++) {
            temp = temp + clusters[i].getSumSqr();
        }
        mSWCSS = temp;
    }
    public void startAnalysis() {
        //set Starting centroid positions - Start of Step 1
        setInitialCentroids();
        int n = 0;
        //assign DataPoint to clusters
        loop1: while (true) {
            for (int l = 0; l < clusters.length; l++)
            {
                clusters[l].addDataPoint((DataPoint)mDataPoints.elementAt(n));
                n++;
                if (n >= mDataPoints.size())
                    break loop1;
            }
        }
        //calculate E for all the clusters
        calcSWCSS();
        //recalculate Cluster centroids - Start of Step 2
        for (int i = 0; i < clusters.length; i++) {
            clusters[i].getCentroid().calcCentroid();
        }
        //recalculate E for all the clusters
        calcSWCSS();
        for (int i = 0; i < miter; i++) {
            //enter the loop for cluster 1
            for (int j = 0; j < clusters.length; j++) {
                for (int k = 0; k < clusters[j].getNumDataPoints(); k++) {
                    //pick the first element of the first cluster
                    //get the current Euclidean distance
                    double tempEuDt = clusters[j].getDataPoint(k).getCurrentEuDt();
                    Cluster tempCluster = null;
                    boolean matchFoundFlag = false;
                    //call testEuclidean distance for all clusters
                    for (int l = 0; l < clusters.length; l++) {
                    //if testEuclidean < currentEuclidean then
                        if (tempEuDt > clusters[j].getDataPoint(k).testEuclideanDistance(clusters[l].getCentroid())) {
                            tempEuDt = clusters[j].getDataPoint(k).testEuclideanDistance(clusters[l].getCentroid());
                            tempCluster = clusters[l];
                            matchFoundFlag = true;
                        }
                        //if statement - Check whether the Last EuDt is > Present EuDt
                        }
//for variable 'l' - Looping between different Clusters for matching a Data Point.
//add DataPoint to the cluster and calcSWCSS
       if (matchFoundFlag) {
                tempCluster.addDataPoint(clusters[j].getDataPoint(k));
                clusters[j].removeDataPoint(clusters[j].getDataPoint(k));
                        for (int m = 0; m < clusters.length; m++) {
                            clusters[m].getCentroid().calcCentroid();
                        }
//for variable 'm' - Recalculating centroids for all Clusters
                        calcSWCSS();
                    }
//if statement - A Data Point is eligible for transfer between Clusters.
                }
                //for variable 'k' - Looping through all Data Points of the current Cluster.
            }//for variable 'j' - Looping through all the Clusters.
        }//for variable 'i' - Number of iterations.
    }
    public Vector[] getClusterOutput() {
        Vector v[] = new Vector[clusters.length];
        for (int i = 0; i < clusters.length; i++) {
            v[i] = clusters[i].getDataPoints();
        }
        return v;
    }

    private void setInitialCentroids() {
        //kn = (round((max-min)/k)*n)+min where n is from 0 to (k-1).
        double cx = 0, cy = 0;
        for (int n = 1; n <= clusters.length; n++) {
            cx = (((getMaxXValue() - getMinXValue()) / (clusters.length + 1)) * n) + getMinXValue();
            cy = (((getMaxYValue() - getMinYValue()) / (clusters.length + 1)) * n) + getMinYValue();
            Centroid c1 = new Centroid(cx, cy);
            clusters[n - 1].setCentroid(c1);
            c1.setCluster(clusters[n - 1]);
        }
    }
    private double getMaxXValue() {
        double temp;
        temp = ((DataPoint) mDataPoints.elementAt(0)).getX();
        for (int i = 0; i < mDataPoints.size(); i++) {
            DataPoint dp = (DataPoint) mDataPoints.elementAt(i);
            temp = (dp.getX() > temp) ? dp.getX() : temp;
        }
        return temp;
    }
    private double getMinXValue() {
        double temp = 0;
        temp = ((DataPoint) mDataPoints.elementAt(0)).getX();
        for (int i = 0; i < mDataPoints.size(); i++) {
            DataPoint dp = (DataPoint) mDataPoints.elementAt(i);
            temp = (dp.getX() < temp) ? dp.getX() : temp;
        }
        return temp;
    }
    private double getMaxYValue() {
        double temp = 0;
        temp = ((DataPoint) mDataPoints.elementAt(0)).getY();
        for (int i = 0; i < mDataPoints.size(); i++) {
            DataPoint dp = (DataPoint) mDataPoints.elementAt(i);
            temp = (dp.getY() > temp) ? dp.getY() : temp;
        }
        return temp;
    }
    private double getMinYValue() {
        double temp