linearclassifier.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 java.text.*;

import edu.stanford.nlp.util.*;
import edu.stanford.nlp.dbm.*;

/** @author Dan Klein */

public class LinearClassifier implements Classifier {

  Counter weights = null;
  Set labels = null;

  Set labels() {
    return labels;
  }

  Set features() {
    Set features = new HashSet();
    for (Iterator i=weights.seenSet().iterator(); i.hasNext();) {
      Pair p = (Pair)i.next();
      features.add(p.first());
    }
    return features;
  }

  public double weight(Object feature, Object label) {
    Pair p = new Pair(feature, label);
    return weights.countOf(p);
  }

  public Counter scoresOf(Datum example) {
    Counter scores = new Counter();
    for (Iterator i=labels().iterator(); i.hasNext();) {
      Object l = i.next();
      double score = 0.0;
      for (Iterator j=example.asFeatures().iterator(); j.hasNext();) {
	Object f = j.next();
	score += weight(f, l);
      }
      scores.setCount(l, score);
    }
    return scores;
  }

  public void dump() {
    Datum allFeatures = new BasicDatum(features(), new String(""));
    justificationOf(allFeatures);
  }

  public void justificationOf(Datum example) {
    int featureLength = 0;
    int labelLength = 6;
    NumberFormat nf = NumberFormat.getNumberInstance();
    nf.setMinimumFractionDigits(2);
    nf.setMaximumFractionDigits(2);
    if (nf instanceof DecimalFormat)
      ((DecimalFormat)nf).setPositivePrefix(" ");
    for (Iterator j=example.asFeatures().iterator(); j.hasNext();) {
      Object f = j.next();
      featureLength = Math.max(featureLength, f.toString().length());
    }
    for (Iterator i=labels().iterator(); i.hasNext();) {
      Object l = i.next();
      labelLength = Math.max(labelLength, l.toString().length());
    }
    StringBuffer header = new StringBuffer("");
    for (int s=0; s<featureLength; s++) {
      header.append(' ');
    }
    for (Iterator i=labels().iterator(); i.hasNext();) {
      Object l = i.next();
      header.append(' ');
      String str = l.toString();
      header.append(str);
      for (int s=str.length(); s<labelLength; s++)
	header.append(' ');
    }
    System.out.println(header);
    for (Iterator j=example.asFeatures().iterator(); j.hasNext();) {
      Object f = j.next();
      String fStr = f.toString();
      StringBuffer line = new StringBuffer(fStr);
      for (int s=fStr.length(); s<featureLength; s++) {
	line.append(' ');
      }
      for (Iterator i=labels().iterator(); i.hasNext();) {
	Object l = i.next();
	String lStr = nf.format(weight(f, l));
	line.append(' ');
	line.append(lStr);
	for (int s=lStr.length(); s<labelLength; s++) {
	  line.append(' ');
	}
      }
      System.out.println(line);
    }
    Counter scores = scoresOf(example);
    StringBuffer footer = new StringBuffer("Total:");
    for (int s=footer.length(); s<featureLength; s++) {
      footer.append(' ');
    }
    for (Iterator i=labels().iterator(); i.hasNext();) {
      Object l = i.next();
      footer.append(' ');
      String str = nf.format(scores.countOf(l));
      footer.append(str);
      for (int s=str.length(); s<labelLength; s++)
	footer.append(' ');
    }
    System.out.println(footer);    
  }

  public Object classOf(Datum example) {
    Counter scores = scoresOf(example);
    return scores.argmax();
  }

  Set extractLabels(Counter c) {
    Set s = new HashSet();
    for (Iterator i=c.seenSet().iterator(); i.hasNext();) {
      Pair p = (Pair)i.next();
      Object l = p.second();
      s.add(l);
    }
    return s;
  }

  public LinearClassifier(Counter weights) {
    this.weights = weights;
    this.labels = extractLabels(weights);
  }
}