/**
* Copyright (c) 2009, Regents of the University of Colorado All rights
* reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer. Redistributions in binary
* form must reproduce the above copyright notice, this list of conditions and
* the following disclaimer in the documentation and/or other materials provided
* with the distribution. Neither the name of the University of Colorado at
* Boulder nor the names of its contributors may be used to endorse or promote
* products derived from this software without specific prior written
* permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
package clear.train.algorithm;
import clear.train.AbstractTrainer;
import clear.train.kernel.AbstractKernel;
import java.util.Arrays;
/**
* RRM algorithm.
*
* @author Jinho D. Choi <b>Last update:</b> 11/5/2010
*/
public class RRM implements IAlgorithm {
private int i_K;
private double d_mu;
private double d_eta;
private double d_c;
public RRM(int K, double mu, double eta, double c) {
i_K = K;
d_mu = mu;
d_eta = eta;
d_c = c;
}
@Override
public double[] getWeight(AbstractKernel kernel, int currLabel) {
double[] pWeight = new double[kernel.D];
Arrays.fill(pWeight, d_mu);
double[] nWeight = new double[kernel.D];
Arrays.fill(nWeight, d_mu);
double[] alpha = new double[kernel.N];
double[] bWeight = new double[kernel.D];
double bestAcc = -1;
int bestK = 0;
double p, min1, min2, min, delta, delta_y_i, currAcc;
byte[] aY = new byte[kernel.N];
byte y_i;
int i, k;
int[] x_i;
for (i = 0; i < kernel.N; i++) {
aY[i] = (kernel.a_ys.get(i) == currLabel) ? (byte) 1 : (byte) -1;
}
for (k = 1; k <= i_K; k++) {
for (i = 0; i < kernel.N; i++) {
// retreive x_i, y_i
x_i = kernel.a_xs.get(i);
y_i = aY[i];
// calculate p
if (kernel.b_binary) {
p = getScore(pWeight, nWeight, x_i) * y_i;
} else {
p = getScore(pWeight, nWeight, x_i, kernel.a_vs.get(i)) * y_i;
}
// calculate delta
min1 = 2 * d_c - alpha[i];
min2 = d_eta * ((d_c - alpha[i]) / d_c - p);
min = Math.min(min1, min2);
delta = Math.max(min, -alpha[i]);
delta_y_i = delta * y_i;
// update weights
pWeight[0] *= Math.exp(delta_y_i);
nWeight[0] *= Math.exp(-delta_y_i);
for (int idx : x_i) {
pWeight[idx] *= Math.exp(delta_y_i);
nWeight[idx] *= Math.exp(-delta_y_i);
}
// update alpha (boosting factor)
alpha[i] += delta;
}
currAcc = getF1Score(kernel, aY, pWeight, nWeight);
if (bestAcc < currAcc) {
setWeight(kernel.D, bWeight, pWeight, nWeight);
bestAcc = currAcc;
bestK = k;
}
if (currAcc == 1) {
break;
}
}
AbstractKernel.normalize(bWeight);
StringBuilder build = new StringBuilder();
build.append("- label = ");
build.append(currLabel);
build.append(": k = ");
build.append(bestK);
build.append(", acc = ");
build.append(bestAcc);
AbstractTrainer.out.println(build.toString());
return (bWeight);
}
/**
* bWeight[i] = pWeight[i] - nWeight[i]
*/
private void setWeight(int D, double[] bWeight, double[] pWeight, double[] nWeight) {
for (int i = 0; i < D; i++) {
bWeight[i] = pWeight[i] - nWeight[i];
}
}
/**
* Returns the score of a training instance
* <code>x</code> using the balanced weight vectors.
*
* @param pWeight positive weight vector
* @param nWeight negative weight vector
* @param x training instance (indices start from 1)
*/
private double getScore(double[] pWeight, double[] nWeight, int[] x) {
double score = pWeight[0] - nWeight[0];
for (int idx : x) {
score += (pWeight[idx] - nWeight[idx]);
}
return score;
}
private double getScore(double[] pWeight, double[] nWeight, int[] x, double[] v) {
double score = pWeight[0] - nWeight[0];
int idx, i;
for (i = 0; i < x.length; i++) {
idx = x[i];
score += (pWeight[idx] - nWeight[idx]) * v[i];
}
return score;
}
/**
* Returns F1 score of the balanced weight vectors.
*
* @param pWeight positive weight vector
* @param nWeight negative weight vector
*/
private double getF1Score(AbstractKernel kernel, byte[] aY, double[] pWeight, double[] nWeight) {
int correct = 0, pTotal = 0, rTotal = 0, i;
byte y_i;
double score;
for (i = 0; i < kernel.N; i++) {
y_i = aY[i];
if (kernel.b_binary) {
score = getScore(pWeight, nWeight, kernel.a_xs.get(i));
} else {
score = getScore(pWeight, nWeight, kernel.a_xs.get(i), kernel.a_vs.get(i));
}
if (score > 0) {
if (y_i == 1) {
correct++;
}
pTotal++;
}
if (y_i == 1) {
rTotal++;
}
}
double precision = (pTotal == 0) ? 0 : (double) correct / pTotal;
double recall = (rTotal == 0) ? 0 : (double) correct / rTotal;
if (precision + recall == 0) {
return 0;
}
return 2 * (precision * recall) / (precision + recall);
}
}