Correlation.java

/**
 * GeDBIT.index.algorithms.Correlation
 * 
 * Copyright Information:
 * 
 * Change Log:
 * 2012.4: Added by Kewei Ma
 */
package GeDBIT.index.algorithms;

import cern.colt.matrix.DoubleMatrix2D;
import cern.colt.matrix.doublealgo.Statistic;

public class Correlation {
    /**
     * Use correlation matrix to do select pivots
     * 
     * @param distance
     *            matrix
     * @param pivots
     *            number
     * @return pivots
     */
    public static int[] runCor(DoubleMatrix2D matrix, int numP) {
	// pivots
	int result[] = new int[numP];
	// whether data point is selected as pivot
	boolean isChosen[] = new boolean[matrix.rows()];
	// correlation matrix
	DoubleMatrix2D coMatrix = Statistic.covariance(matrix);

	// choose the first pivot with largest variance
	for (int i = 1; i < coMatrix.rows(); i++) {
	    if (coMatrix.get(i, i) > coMatrix.get(result[0], result[0])) {
		result[0] = i;
	    }
	}
	// set it true
	isChosen[result[0]] = true;

	// select rest pivots
	if (numP > 1) {
	    Statistic.correlation(coMatrix);
	    for (int i = 1; i < numP; i++) {
		double minCor[] = new double[i];
		int minCorIndics[] = new int[i];
		for (int j = 0; j < i; j++) {
		    for (int k = 0; k < coMatrix.rows(); k++) {
			if (isChosen[k] == false) {
			    if (Math.abs(coMatrix.get(k, result[j])) < Math
				    .abs(coMatrix.get(minCorIndics[j],
					    result[j]))) {
				minCorIndics[j] = k;
				minCor[j] = Math
					.abs(coMatrix.get(k, result[j]));
			    }
			}
		    }
		}

		// pick largest ones as pivots
		int index = 0;
		for (int j = 1; j < minCorIndics.length; j++) {

		    if (minCor[j] > minCor[index])
			index = j;
		}
		result[i] = minCorIndics[index];
		isChosen[minCorIndics[index]] = true;
	    }
	}

	return result;

    }
}