cglinearclassifierfactory.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.util.*;
import edu.stanford.nlp.optimization.*;


/** @author Dan Klein */

public class CGLinearClassifierFactory extends AbstractLinearClassifierFactory {

  boolean useSum = false;
  double TOL = 1e-4;
  double sigma = 1.0;

  public double[][] trainWeights(int[][] data, int[] labels) {
    Minimizer minimizer = new CGMinimizer(false);
    DiffFunction objective = (useSum ?
			      new SumConditionalObjectiveFunction(numFeatures(), numClasses(), data, labels):
			      new LogConditionalObjectiveFunction(numFeatures(), numClasses(), data, labels, sigma));
    double[] weights = minimizer.minimize(objective, TOL, ((HasInitial)objective).initial());
    //System.err.println("Final derivative norm2: "+((AbstractCachingDiffFunction)objective).norm2(((AbstractCachingDiffFunction)objective).derivative));
    return ((LogConditionalObjectiveFunction)objective).to2D(weights);
  }

  public CGLinearClassifierFactory() {
  };  
  public CGLinearClassifierFactory(boolean useSum) {
    this.useSum = useSum;
  };
  public CGLinearClassifierFactory(double tol) {
    this.TOL = tol;
  };
  public CGLinearClassifierFactory(double tol, boolean useSum) {
    this.TOL = tol;
    this.useSum = useSum;
  };
  public CGLinearClassifierFactory(double tol, boolean useSum, double sigma) {
    this.TOL = tol;
    this.useSum = useSum;
    this.sigma = sigma;
  };
}