logistic.java

MacroWeka扩展了著名数据挖掘工具weka

/*
 *    This program is free software; you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation; either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/*
 *    Logistic.java
 *    Copyright (C) 2003 Xin Xu
 *
 */

package weka.classifiers.functions;

import weka.classifiers.*;
import java.util.*;
import java.io.*;
import weka.core.*;
import weka.filters.*;
import weka.filters.unsupervised.attribute.*;

/**
 * Second implementation for building and using a multinomial logistic
 * regression model with a ridge estimator.  <p>
 * 
 * There are some modifications, however, compared to the paper of le
 * Cessie and van Houwelingen(1992): <br>
 *
 * If there are k classes for n instances with m attributes, the
 * parameter matrix B to be calculated will be an m*(k-1) matrix.<br>
 *
 * The probability for class j except the last class is <br>
 * Pj(Xi) = exp(XiBj)/((sum[j=1..(k-1)]exp(Xi*Bj))+1) <br>
 * The last class has probability <br>
 * 1-(sum[j=1..(k-1)]Pj(Xi)) = 1/((sum[j=1..(k-1)]exp(Xi*Bj))+1) <br>
 *
 * The (negative) multinomial log-likelihood is thus: <br>
 * L = -sum[i=1..n]{
 * sum[j=1..(k-1)](Yij * ln(Pj(Xi))) +
 * (1 - (sum[j=1..(k-1)]Yij)) * ln(1 - sum[j=1..(k-1)]Pj(Xi))
 * } + ridge * (B^2) <br>
 *
 * In order to find the matrix B for which L is minimised, a
 * Quasi-Newton Method is used to search for the optimized values of
 * the m*(k-1) variables.  Note that before we use the optimization
 * procedure, we "squeeze" the matrix B into a m*(k-1) vector.  For
 * details of the optimization procedure, please check
 * weka.core.Optimization class. <p>
 *
 * Although original Logistic Regression does not deal with instance
 * weights, we modify the algorithm a little bit to handle the
 * instance weights. <p>
 *
 * Reference: le Cessie, S. and van Houwelingen, J.C. (1992). <i>
 * Ridge Estimators in Logistic Regression.</i> Applied Statistics,
 * Vol. 41, No. 1, pp. 191-201. <p>
 *
 * Missing values are replaced using a ReplaceMissingValuesFilter, and
 * nominal attributes are transformed into numeric attributes using a
 * NominalToBinaryFilter.<p>
 *
 * Valid options are:<p>
 *
 * -D <br>
 * Turn on debugging output.<p>
 *
 * -R <ridge> <br>
 * Set the ridge parameter for the log-likelihood.<p>
 * 
 * -M <number of iterations> <br> Set the maximum number of iterations
 * (default -1, iterates until convergence).<p>
 *
 * @author Xin Xu ([e-mail])
 * @version $Revision: 1.1 $ */
public class Logistic extends Classifier 
  implements OptionHandler, WeightedInstancesHandler {
  
  /** The coefficients (optimized parameters) of the model */
  protected double [][] m_Par;
    
  /** The data saved as a matrix */
  protected double [][] m_Data;
    
  /** The number of attributes in the model */
  protected int m_NumPredictors;
    
  /** The index of the class attribute */
  protected int m_ClassIndex;
    
  /** The number of the class labels */
  protected int m_NumClasses;
    
  /** The ridge parameter. */
  protected double m_Ridge = 1e-8;
    
  /* An attribute filter */
  private RemoveUseless m_AttFilter;
    
  /** The filter used to make attributes numeric. */
  private NominalToBinary m_NominalToBinary;
    
  /** The filter used to get rid of missing values. */
  private ReplaceMissingValues m_ReplaceMissingValues;
    
  /** Debugging output */
  protected boolean m_Debug;

  /** Log-likelihood of the searched model */
  protected double m_LL;
    
  /** The maximum number of iterations. */
  private int m_MaxIts = -1;
    
  /**
   * Returns a string describing this classifier
   * @return a description of the classifier suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {
    return "Class for building and using a multinomial logistic "
      +"regression model with a ridge estimator.\n\n"
      +"There are some modifications, however, compared to the paper of "
      +"leCessie and van Houwelingen(1992): \n\n" 
      +"If there are k classes for n instances with m attributes, the "
      +"parameter matrix B to be calculated will be an m*(k-1) matrix.\n\n"
      +"The probability for class j with the exception of the last class is\n\n"
      +"Pj(Xi) = exp(XiBj)/((sum[j=1..(k-1)]exp(Xi*Bj))+1) \n\n"
      +"The last class has probability\n\n"
      +"1-(sum[j=1..(k-1)]Pj(Xi)) \n\t= 1/((sum[j=1..(k-1)]exp(Xi*Bj))+1)\n\n"
      +"The (negative) multinomial log-likelihood is thus: \n\n"
      +"L = -sum[i=1..n]{\n\tsum[j=1..(k-1)](Yij * ln(Pj(Xi)))"
      +"\n\t+(1 - (sum[j=1..(k-1)]Yij)) \n\t* ln(1 - sum[j=1..(k-1)]Pj(Xi))"
      +"\n\t} + ridge * (B^2)\n\n"
      +"In order to find the matrix B for which L is minimised, a "
      +"Quasi-Newton Method is used to search for the optimized values of "
      +"the m*(k-1) variables.  Note that before we use the optimization "
      +"procedure, we 'squeeze' the matrix B into a m*(k-1) vector.  For "
      +"details of the optimization procedure, please check "
      +"weka.core.Optimization class.\n\n"
      +"Although original Logistic Regression does not deal with instance "
      +"weights, we modify the algorithm a little bit to handle the "
      +"instance weights.\n\n"
      +"For more information see:\n\nle Cessie, S. and van Houwelingen, J.C. (1992). " 
      +"Ridge Estimators in Logistic Regression.  Applied Statistics, "
      +"Vol. 41, No. 1, pp. 191-201. \n\n"
      +"Note: Missing values are replaced using a ReplaceMissingValuesFilter, and "
      +"nominal attributes are transformed into numeric attributes using a "
      +"NominalToBinaryFilter.";
  }

  /**
   * Returns an enumeration describing the available options
   *
   * @return an enumeration of all the available options
   */
  public Enumeration listOptions() {
    Vector newVector = new Vector(3);
    newVector.addElement(new Option("\tTurn on debugging output.",
				    "D", 0, "-D"));
    newVector.addElement(new Option("\tSet the ridge in the log-likelihood.",
				    "R", 1, "-R <ridge>"));
    newVector.addElement(new Option("\tSet the maximum number of iterations"+
				    " (default -1, until convergence).",
				    "M", 1, "-M <number>"));
    return newVector.elements();
  }
    
  /**
   * Parses a given list of options. Valid options are:<p>
   *
   * -D <br>
   * Turn on debugging output.<p>
   *
   * -R ridge <br>
   * Set the ridge parameter for the log-likelihood.<p>
   * 
   * -M num <br>
   * Set the maximum number of iterations.
   * (default -1, until convergence)<p>
   *
   * @param options the list of options as an array of strings
   * @exception Exception if an option is not supported
   */
  public void setOptions(String[] options) throws Exception {
    setDebug(Utils.getFlag('D', options));

    String ridgeString = Utils.getOption('R', options);
    if (ridgeString.length() != 0) 
      m_Ridge = Double.parseDouble(ridgeString);
    else 
      m_Ridge = 1.0e-8;
	
    String maxItsString = Utils.getOption('M', options);
    if (maxItsString.length() != 0) 
      m_MaxIts = Integer.parseInt(maxItsString);
    else 
      m_MaxIts = -1;
  }
    
  /**
   * Gets the current settings of the classifier.
   *
   * @return an array of strings suitable for passing to setOptions
   */
  public String [] getOptions() {
	
    String [] options = new String [5];
    int current = 0;
	
    if (getDebug()) 
      options[current++] = "-D";
    options[current++] = "-R";
    options[current++] = ""+m_Ridge;	
    options[current++] = "-M";
    options[current++] = ""+m_MaxIts;
    while (current < options.length) 
      options[current++] = "";
    return options;
  }
   
  /**
   * Returns the tip text for this property
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String debugTipText() {
    return "Output debug information to the console.";
  }

  /**
   * Sets whether debugging output will be printed.
   *
   * @param debug true if debugging output should be printed
   */
  public void setDebug(boolean debug) {
    m_Debug = debug;
  }
    
  /**
   * Gets whether debugging output will be printed.
   *
   * @return true if debugging output will be printed
   */
  public boolean getDebug() {
    return m_Debug;
  }      

  /**
   * Returns the tip text for this property
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String ridgeTipText() {
    return "Set the Ridge value in the log-likelihood.";
  }

  /**
   * Sets the ridge in the log-likelihood.
   *
   * @param ridge the ridge
   */
  public void setRidge(double ridge) {
    m_Ridge = ridge;
  }
    
  /**
   * Gets the ridge in the log-likelihood.
   *
   * @return the ridge
   */
  public double getRidge() {
    return m_Ridge;
  }
   
  /**
   * Returns the tip text for this property
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String maxItsTipText() {
    return "Maximum number of iterations to perform.";
  }

  /**
   * Get the value of MaxIts.
   *
   * @return Value of MaxIts.
   */
  public int getMaxIts() {
	
    return m_MaxIts;
  }
    
  /**
   * Set the value of MaxIts.
   *
   * @param newMaxIts Value to assign to MaxIts.
   */
  public void setMaxIts(int newMaxIts) {
	
    m_MaxIts = newMaxIts;
  }    
    
  private class OptEng extends Optimization{
    // Weights of instances in the data
    private double[] weights;

    // Class labels of instances
    private int[] cls;
	
    /* Set the weights of instances
     * @param d the weights to be set
     */ 
    public void setWeights(double[] w) {
      weights = w;
    }
	
    /* Set the class labels of instances
     * @param d the class labels to be set
     */ 
    public void setClassLabels(int[] c) {
      cls = c;
    }
	
    /** 
     * Evaluate objective function
     * @param x the current values of variables
     * @return the value of the objective function 
     */
    protected double objectiveFunction(double[] x){
      double nll = 0; // -LogLikelihood
      int dim = m_NumPredictors+1; // Number of variables per class
	    
      for(int i=0; i<cls.length; i++){ // ith instance

	double[] exp = new double[m_NumClasses-1];
	int index;
	for(int offset=0; offset<m_NumClasses-1; offset++){ 
	  index = offset * dim;
	  for(int j=0; j<dim; j++)
	    exp[offset] += m_Data[i][j]*x[index + j];
	}
	double max = exp[Utils.maxIndex(exp)];
	double denom = Math.exp(-max);
	double num;
	if (cls[i] == m_NumClasses - 1) { // Class of this instance
	  num = -max;
	} else {
	  num = exp[cls[i]] - max;
	}
	for(int offset=0; offset<m_NumClasses-1; offset++){