sumconditionalobjectivefunction.java

Standord Classifier实现了一个基于Java的最大熵分类器。用于模式识别

package edu.stanford.nlp.classify;

import java.util.*;

import edu.stanford.nlp.optimization.*;

/** @author Dan Klein */

public class SumConditionalObjectiveFunction extends LogConditionalObjectiveFunction {

  protected void calculate(double[] x) {
    //System.out.println("Checking at: "+x[0]+" "+x[1]+" "+x[2]);
    value = 0.0;
    Arrays.fill(derivative, 0.0);
    double[] sums = new double[numClasses];
    double[] probs = new double[numClasses];
    double[] counts = new double[numClasses];
    Arrays.fill(counts, 0.0);
    for (int d=0; d<data.length; d++) {
      int[] features = data[d];
      // activation
      Arrays.fill(sums, 0.0);
      for (int c=0; c<numClasses; c++) {
	for (int f=0; f<features.length; f++) {
	  int i = indexOf(features[f], c);
	  sums[c] += x[i];
	}
      }
      // expectation
      double total = Double.NEGATIVE_INFINITY;
      for (int c=0; c<numClasses; c++) {
	total = addLPs(total, sums[c]);
      }
      int ld = labels[d];
      for (int c=0; c<numClasses; c++) {
	probs[c] = Math.exp(sums[c]-total);
	for (int f=0; f<features.length; f++) {
	  int i = indexOf(features[f], c);
	  derivative[i] += probs[ld]*probs[c];
	}
      }
      // observed
      for (int f=0; f<features.length; f++) {
	int i = indexOf(features[f], labels[d]);
	derivative[i] -= probs[ld];
      }
      value -= probs[ld];
    }
    // priors
    if (true) {
      for (int i=0; i<x.length; i++) {
	double k = 1.0;
	double w = x[i];
	value += k*w*w/2.0;
	derivative[i] += k*w;
      }
    }
  }

  public SumConditionalObjectiveFunction(int numFeatures, int numClasses, int[][] data, int[] labels) {
    super(numFeatures, numClasses, data, labels);
  }


}