exampleset.java

来自「Boosting算法软件包」· Java 代码 · 共 277 行

/* -*- mode: java; c-basic-offset: 2; indent-tabs-mode: nil -*- */
package jboost.examples;

//import java.io.*;
import java.util.ArrayList;

import jboost.Predictor;
import jboost.booster.Prediction;
import jboost.learner.IncompAttException;
import jboost.monitor.Monitor;

/**
 * Holds a set of examples and the text that describes them 
 * @author Yoav Freund
 * @version $Header: /cvsroot/jboost/jboost/src/jboost/examples/ExampleSet.java,v 1.2 2007/09/18 03:25:59 aarvey Exp $
 */
public class ExampleSet {
  /** 
   * The content of the ExampleSet */
  private Example[] exampleSet; 
  
  /** 
   * A temporary storage for the examples
   */
  private ArrayList exampleList;
  private int nextExample;	// an internal counter for checking that the
  // indexes are coming in order
  /**
   * The textual description of the fields in an Example */
  private ExampleDescription exampleDescription;
  private int noOfLabels=2;	// no. of possible m_labels
  private boolean multiLabel=false;	// true if multiple m_labels allowed
  private boolean isBinary=true;	// identifies the data as binary-labeled
  
  /** last iteration on which calc* was called */
  private int lastIter = -2;  
  private Prediction[] prediction = null;

  /** default constructor */
  public ExampleSet(){
    exampleSet=null;
    exampleDescription=null;
    exampleList = new ArrayList();
    nextExample = 0;
  }
  /** constructor that gets an exampleDescription */
  public ExampleSet(ExampleDescription exampleDescription) {
    this();
    this.exampleDescription = exampleDescription;
    try {
      noOfLabels = exampleDescription.getLabelDescription().getNoOfValues();
      multiLabel = exampleDescription.getLabelDescription().isMultiLabel();
    }
    catch (IncompAttException e) {
      throw new 
      RuntimeException("While initializing ExampleSet," +
      "LabelDescription does not support getNoOfValues\n");
    }
    isBinary = (noOfLabels==2 && !multiLabel);
  }
  /** add an example */
  public void addExample(int index, Example example) {
    if(index != nextExample) 
      throw new RuntimeException("ExampleSet.addExample: index="+index+
          " Expected next index="+nextExample);
    example.setDescription(exampleDescription);
    exampleList.add(example);
    nextExample++;
  }
  /** finalize the dataset */
  public void finalizeData() {
    exampleSet = (Example[]) exampleList.toArray(new Example[0]);
    prediction = new Prediction[exampleSet.length];
    exampleList.clear();	// free the space
    exampleList=null;
  }
  
  /** get no of examples */
  public int getExampleNo() {
    if(exampleSet == null) return exampleList.size();
    else return exampleSet.length;
  }

  /** get no of examples */
  public int size() {
      return getExampleNo();
  }
  
  /** get example no. i */
  public Example getExample(int i) { 
    if(exampleSet == null) return (Example) exampleList.get(i);
    else return exampleSet[i];
  }
  
  /** calculate error of a Predictor on this ExampleSet.
   * If curIter equals iteration number from last call to
   * calcError, calcMargins or calcScores, and base != null then
   * previously computed predictions are used.  If curIter equals
   * last iteration number plus one and base != null, then base
   * predictor is used to update previously computed predictions.
   * If base predictor is null and in all other cases, combined
   * predictor is used for predictions.
   *  @param curIter current iteration number
   *  @param combined current combined predictor
   *  @param base   last added base predictor
   */
  public double calcError(int curIter, Predictor combined, Predictor base) {
    updatePredictions(curIter, combined, base);
    int size = exampleSet.length;
    if(size == 0) {
      return Double.NaN;
      //	    throw new RuntimeException("ExampleSet.calcError: ExampleSet is empty");
    }
    double errors=0.0;
    int i=0;
    try{
      for(i=0; i<size; i++) {
        Example x = exampleSet[i];
        errors += x.getLabel().
        lossOfPrediction(prediction[i].getBestClass());
      }
    }
    catch (Exception e) {
      if(Monitor.logLevel>3) Monitor.log("in ExampleSet.calcError():" +
          " got exception on example no "+i);
      if(Monitor.logLevel>3) Monitor.log(e.getMessage());
      e.printStackTrace();
    }
    
    
    return errors / size;
  }
  
  /** calculate m_margins of a Predictor on this ExampleSet
   * see parameter description at calcError
   */
  public ArrayList calcMargins(int curIter, Predictor combined, Predictor base) {
    updatePredictions(curIter, combined, base);
    int size = exampleSet.length;
    if(size == 0) return null;
    
    ArrayList margins = new ArrayList();
    
    int i=0;
    try{
      for(i=0; i<size; i++) {
        Example x = exampleSet[i];
        double[] tmp = prediction[i].getMargins(x.getLabel());
        margins.add(tmp);
      }
    }
    catch (Exception e) {
      if(Monitor.logLevel>3) Monitor.log("in ExampleSet.calcMargins():" +
          " got exception on example no "+i);
      if(Monitor.logLevel>3) Monitor.log(e.getMessage());
      e.printStackTrace();
    }
    
    return margins;
  }
  
  /** calculate scores of a Predictor on this ExampleSet
   * see parameter description at calcError
   */
  public ArrayList calcScores(int curIter, Predictor combined, Predictor base) {
    updatePredictions(curIter, combined, base);
    int size = exampleSet.length;
    if(size == 0) return null;
    
    ArrayList scores = new ArrayList();
    
    int i=0;
    double[] tmp = null;
    double tmp0 = 0;
    try{
      for(i=0; i<size; i++) {
        tmp = prediction[i].getClassScores();
        if(isBinary) { // for binary m_labels, keep only one score
          tmp0 = tmp[0];
          tmp = new double[1];
          tmp[0] = tmp0;
        }
        scores.add(tmp);
      }
    }
    catch (Exception e) {
      if(Monitor.logLevel>3) Monitor.log("in ExampleSet.calcScores():" +
          " got exception on example no "+i);
      if(Monitor.logLevel>3) Monitor.log(e.getMessage());
      e.printStackTrace();
    }
    
    return scores;
  }
  


  
  /** updates the saved predictions
   * see parameter description at calcError
   */
  private void updatePredictions(int curIter,
      Predictor combined,
      Predictor base) {


    if (base != null) {
      if (curIter == lastIter)
        return;
      if (curIter == lastIter + 1) {
        for (int i = 0; i < exampleSet.length; i++)
          prediction[i].add
          (base.predict(exampleSet[i].getInstance()));
        lastIter = curIter;
        return;
      }
    }


    for (int i = 0; i < exampleSet.length; i++)
      prediction[i] =
        combined.predict(exampleSet[i].getInstance(),curIter);

    lastIter = curIter;
  }
  
  
  
  
  /** get the binary-m_labels equivalent of the m_labels in this ExampleSet */
  public ArrayList getBinaryLabels() {
    int size = exampleSet.length;
    if(size == 0) return null;
    
    ArrayList labels = new ArrayList();
    
    int i=0;
    Boolean[] tmp = null; 
    try{
      for(i=0; i<size; i++) {
        Label l = exampleSet[i].getLabel();
        if(isBinary) {
          tmp = new Boolean[1];
          tmp[0] = new Boolean(l.getMultiValue(0));
        } else { 
          tmp = new Boolean[noOfLabels];
          for(int j=0; j<noOfLabels; j++) {
            tmp[j] = new Boolean(l.getMultiValue(j));
          }
	}
       labels.add(tmp);
      }
    }
    catch (Exception e) {
      if(Monitor.logLevel>3) Monitor.log("in ExampleSet.getBinaryLabels():" +
          " got exception on example no "+i);
      if(Monitor.logLevel>3) Monitor.log(e.getMessage());
      e.printStackTrace();
    }
    
    return labels;
  }
  
  /** print in human-readable format */
  public String toString() {
    String s="ExampleSet of "+exampleSet.length+" examples\n";
    s += "exampleDescription\n"+exampleDescription;
    for(int i=0; i<exampleSet.length; i++) 
      s += i+"\t"+exampleSet[i];
    return s;
  }
}