PearsonSigWeightingMetric.java

/**
 * Compute the Pearson similarity between profiles - also applies significance weighting
 * 
 * Michael O'Mahony
 * 20/01/2011
 */

package similarity.metric;

import java.util.Set;

import profile.Profile;

public class PearsonSigWeightingMetric implements SimilarityMetric
{
	private int N; // significance weighting threshold value
	
	/**
	 * constructor - creates a new PearsonSigWeightingMetric object
	 * @param N - significance weighting threshold value
	 */
	public PearsonSigWeightingMetric(final int N)
	{
		this.N = N;
	}

	/**
	 * computes the similarity between profiles
	 * @param profile 1 
	 * @param profile 2
	 */
	public double getSimilarity(final Profile p1, final Profile p2)
	{
		double sum_r1 = 0;
		double sum_r1_sq = 0;
		double sum_r2 = 0;
		double sum_r2_sq = 0;
		double sum_r1_r2 = 0;

		Set<Integer> common = p1.getCommonIds(p2);
		for(Integer id: common)
		{
			double r1 = p1.getValue(id).doubleValue();
			double r2 = p2.getValue(id).doubleValue();

			sum_r1 += r1;
			sum_r1_sq += r1 * r1;
			sum_r2 += r2;
			sum_r2_sq += r2 * r2;
			sum_r1_r2 += r1 * r2;
		}

		double above = (common.size() > 0) ? sum_r1_r2 - (sum_r1 * sum_r2) / common.size() : 0;
		double below = (common.size() > 0) ? Math.sqrt( (sum_r1_sq - (sum_r1 * sum_r1) / common.size()) * (sum_r2_sq - (sum_r2 * sum_r2) / common.size()) ) : 0;
		double sim = (below > 0) ? above / below : 0;
		return (common.size() < N) ? sim * (common.size() / N) : sim;
	}
}