nnge.java

代码是一个分类器的实现,其中使用了部分weka的源代码。可以将项目导入eclipse运行

	   m_MI_MaxArray[attrIndex] < inst.value(attrIndex) || 
	   inst.value(attrIndex) < m_MI_MinArray[attrIndex]){

	  /* then update them */
	  if(Double.isNaN(m_MI_MaxArray[attrIndex])) m_MI_MaxArray[attrIndex] = inst.value(attrIndex);
	  if(Double.isNaN(m_MI_MinArray[attrIndex])) m_MI_MinArray[attrIndex] = inst.value(attrIndex);
	  if(m_MI_MaxArray[attrIndex] < inst.value(attrIndex)) m_MI_MaxArray[attrIndex] = inst.value(attrIndex);
	  if(m_MI_MinArray[attrIndex] > inst.value(attrIndex)) m_MI_MinArray[attrIndex] = inst.value(attrIndex);
		    
	  /* and re-compute everything from scratch... (just for this attribute) */
	  double delta = (m_MI_MaxArray[attrIndex] - m_MI_MinArray[attrIndex]) / (double) m_NumFoldersMI;

	  /* for each interval */
	  for(int inter = 0; inter < m_NumFoldersMI; inter++){

	    m_MI_NumAttrInter[attrIndex][inter] = 0;

	    /* for each class */
	    for(int cclass = 0; cclass < m_Train.numClasses(); cclass++){
			    
	      m_MI_NumAttrClassInter[attrIndex][cclass][inter] = 0;

	      /* count */
	      Enumeration enu = m_Train.enumerateInstances();
	      while(enu.hasMoreElements()){
		Instance cur = (Instance) enu.nextElement();
		if(( (m_MI_MinArray[attrIndex] + inter * delta) <= cur.value(attrIndex)       ) &&
		   ( cur.value(attrIndex) <= (m_MI_MinArray[attrIndex] + (inter + 1) * delta) ) &&
		   ( cur.classValue() == cclass ) ){
		  m_MI_NumAttrInter[attrIndex][inter]++;
		  m_MI_NumAttrClassInter[attrIndex][cclass][inter]++;
		}
	      }
	    }
	  }
		
	  /* max-min don't have to be updated */
	} else {

	  /* still have to incr the card of the correct interval */
	  double delta = (m_MI_MaxArray[attrIndex] - m_MI_MinArray[attrIndex]) / (double) m_NumFoldersMI;
		    
	  /* for each interval */
	  for(int inter = 0; inter < m_NumFoldersMI; inter++){
	    /* which contains inst*/
	    if(( (m_MI_MinArray[attrIndex] + inter * delta) <= inst.value(attrIndex)       ) &&
	       ( inst.value(attrIndex) <= (m_MI_MinArray[attrIndex] + (inter + 1) * delta) )){
	      m_MI_NumAttrInter[attrIndex][inter]++;
	      m_MI_NumAttrClassInter[attrIndex][(int) inst.classValue()][inter]++;
	    }
	  }
	}
		
	/* update the mutual information of this attribute... */
	m_MI[attrIndex] = 0;
		
	/* for each interval, for each class */
	for(int inter = 0; inter < m_NumFoldersMI; inter++){
	  for(int cclass = 0; cclass < m_Train.numClasses(); cclass++){
	    double pXY = ((double) m_MI_NumAttrClassInter[attrIndex][cclass][inter]) / ((double) m_MI_NumInst);
	    double pX = ((double) m_MI_NumClass[cclass]) / ((double) m_MI_NumInst);
	    double pY = ((double) m_MI_NumAttrInter[attrIndex][inter]) / ((double) m_MI_NumInst);

	    if(pXY != 0)
	      m_MI[attrIndex] += pXY * Utils.log2(pXY / (pX * pY));
	  }
	}
		
	/* which is a nominal attribute */
      } else if (m_Train.attribute(attrIndex).isNominal()){
		
	/*incr the card of the correct 'values' */
	m_MI_NumAttrValue[attrIndex][(int) inst.value(attrIndex)]++;
	m_MI_NumAttrClassValue[attrIndex][(int) inst.classValue()][(int) inst.value(attrIndex)]++;
		
	/* update the mutual information of this attribute... */
	m_MI[attrIndex] = 0;
		
	/* for each nominal value, for each class */
	for(int attrValue = 0; attrValue < m_Train.attribute(attrIndex).numValues() + 1; attrValue++){
	  for(int cclass = 0; cclass < m_Train.numClasses(); cclass++){
	    double pXY = ((double) m_MI_NumAttrClassValue[attrIndex][cclass][attrValue]) / ((double) m_MI_NumInst);
	    double pX = ((double) m_MI_NumClass[cclass]) / ((double) m_MI_NumInst);
	    double pY = ((double) m_MI_NumAttrValue[attrIndex][attrValue]) / ((double) m_MI_NumInst);
	    if(pXY != 0)
	      m_MI[attrIndex] += pXY * Utils.log2(pXY / (pX * pY));
	  }
	}

	/* not a nominal attribute, not a numeric attribute */
      } else {
	throw new Exception("NNge.updateMI : Cannot deal with 'string attribute'.");
      }
    }	
  }


  /**
   * Init the weight of ex
   * Watch out ! ex shouldn't be in NNge's lists when initialized
   *
   * @param ex the Exemplar to initialise
   */
  private void initWeight(Exemplar ex) {	
    int pos = 0, neg = 0, n = 0;
    Exemplar cur = m_Exemplars;
    if (cur == null){
      ex.setPositiveCount(1);
      ex.setNegativeCount(0);
      return;
    }
    while(cur != null){
      pos += cur.getPositiveCount();
      neg += cur.getNegativeCount();
      n++;
      cur = cur.next;
    }
    ex.setPositiveCount(pos / n);
    ex.setNegativeCount(neg / n);
  }


  /**
   * Adds an Exemplar in NNge's lists
   * Ensure that the exemplar is not already in a list : the links would be broken...
   *
   * @param ex a new Exemplar to add
   */
  private void addExemplar(Exemplar ex) {
	
    /* add ex at the top of the general list */
    ex.next = m_Exemplars;
    if(m_Exemplars != null)
      m_Exemplars.previous = ex;
    ex.previous = null;
    m_Exemplars = ex;

    /* add ex at the top of the corresponding class list */
    ex.nextWithClass = m_ExemplarsByClass[(int) ex.classValue()];
    if(m_ExemplarsByClass[(int) ex.classValue()] != null)
      m_ExemplarsByClass[(int) ex.classValue()].previousWithClass = ex;
    ex.previousWithClass = null;
    m_ExemplarsByClass[(int) ex.classValue()] = ex;
  }


  /**
   * Removes an Exemplar from NNge's lists
   * Ensure that the Exemplar is actually in NNge's lists. 
   *   Likely to do something wrong if this condition is not respected.
   * Due to the list implementation, the Exemplar can appear only once in the lists : 
   *   once removed, the exemplar is not in the lists anymore.
   *
   * @param ex a new Exemplar to add
   */
  private void removeExemplar(Exemplar ex){

    /* remove from the general list */
    if(m_Exemplars == ex){
      m_Exemplars = ex.next;
      if(m_Exemplars != null)
	m_Exemplars.previous = null;
	
    } else {
      ex.previous.next = ex.next;
      if(ex.next != null){
	ex.next.previous = ex.previous;
      }
    }
    ex.next = ex.previous = null;

    /* remove from the class list */
    if(m_ExemplarsByClass[(int) ex.classValue()] == ex){
      m_ExemplarsByClass[(int) ex.classValue()] = ex.nextWithClass;
      if(m_ExemplarsByClass[(int) ex.classValue()] != null)
	m_ExemplarsByClass[(int) ex.classValue()].previousWithClass = null;
	
    } else {
      ex.previousWithClass.nextWithClass = ex.nextWithClass;
      if(ex.nextWithClass != null){
	ex.nextWithClass.previousWithClass = ex.previousWithClass;
      }
    }
    ex.nextWithClass = ex.previousWithClass = null;
  }


  /**
   * returns the weight of indexth attribute
   *
   * @param index attribute's index
   * @return the weight of indexth attribute
   */
  private double attrWeight (int index) {
    return m_MI[index];
  }

    
  /**
   * Returns a description of this classifier.
   *
   * @return a description of this classifier as a string.
   */
  public String toString(){

    String s;
    Exemplar cur = m_Exemplars;
    int i;	

   if (m_MinArray == null) {
      return "No classifier built";
    }
     int[] nbHypClass = new int[m_Train.numClasses()];
    int[] nbSingleClass = new int[m_Train.numClasses()];
    for(i = 0; i<nbHypClass.length; i++){
      nbHypClass[i] = 0;
      nbSingleClass[i] = 0;
    }
    int nbHyp = 0, nbSingle = 0;

    s = "\nNNGE classifier\n\nRules generated :\n";

    while(cur != null){
      s += "\tclass " + m_Train.attribute(m_Train.classIndex()).value((int) cur.classValue()) + " IF : ";
      s += cur.toRules() + "\n";
      nbHyp++;
      nbHypClass[(int) cur.classValue()]++;	    
      if (cur.numInstances() == 1){
	nbSingle++;
	nbSingleClass[(int) cur.classValue()]++;
      }
      cur = cur.next;
    }
    s += "\nStat :\n";
    for(i = 0; i<nbHypClass.length; i++){
      s += "\tclass " + m_Train.attribute(m_Train.classIndex()).value(i) + 
	" : " + Integer.toString(nbHypClass[i]) + " exemplar(s) including " + 
	Integer.toString(nbHypClass[i] - nbSingleClass[i]) + " Hyperrectangle(s) and " +
	Integer.toString(nbSingleClass[i]) + " Single(s).\n";
    }
    s += "\n\tTotal : " + Integer.toString(nbHyp) + " exemplars(s) including " + 
      Integer.toString(nbHyp - nbSingle) + " Hyperrectangle(s) and " +
      Integer.toString(nbSingle) + " Single(s).\n";
	
    s += "\n";
	
    s += "\tFeature weights : ";

    String space = "[";
    for(int ii = 0; ii < m_Train.numAttributes(); ii++){
      if(ii != m_Train.classIndex()){
	s += space + Double.toString(attrWeight(ii));
	space = " ";
      }
    }
    s += "]";
    s += "\n\n";
    return s;
  }



  /** OPTION HANDLER FUNCTION */
    

  /**
   * Returns an enumeration of all the available options..
   *
   * @return an enumeration of all available options.
   */
  public Enumeration listOptions(){

    Vector newVector = new Vector(2);

    newVector.addElement(new Option(
				    "\tNumber of attempts of generalisation.\n",
				    "G", 
				    1, 
				    "-G <value>"));
    newVector.addElement(new Option(
				    "\tNumber of folder for computing the mutual information.\n",
				    "I", 
				    1, 
				    "-I <value>"));

    return newVector.elements();
  }

    
  /**
   * Sets the OptionHandler's options using the given list. All options
   * will be set (or reset) during this call (i.e. incremental setting
   * of options is not possible). <p/>
   * 
   <!-- options-start -->
   * Valid options are: <p/>
   * 
   * <pre> -G &lt;value&gt;
   *  Number of attempts of generalisation.
   * </pre>
   * 
   * <pre> -I &lt;value&gt;
   *  Number of folder for computing the mutual information.
   * </pre>
   * 
   <!-- options-end -->
   *
   * @param options the list of options as an array of strings
   * @throws Exception if an option is not supported
   */
  public void setOptions(String[] options) throws Exception {

    String str;

    /* Number max of attempts of generalisation */
    str = Utils.getOption('G', options);
    if(str.length() != 0){
      m_NumAttemptsOfGene = Integer.parseInt(str);
      if(m_NumAttemptsOfGene < 1)
	throw new Exception("NNge.setOptions : G option's value must be greater than 1.");
    } else {
      m_NumAttemptsOfGene = 5;
    }

    /* Number of folder for computing the mutual information */
    str = Utils.getOption('I', options);
    if(str.length() != 0){
      m_NumFoldersMI = Integer.parseInt(str);
      if(m_NumFoldersMI < 1)
	throw new Exception("NNge.setOptions : I option's value must be greater than 1.");
    } else {
      m_NumFoldersMI = 5;
    }
  }

    
  /**
   * Gets the current option settings for the OptionHandler.
   *
   * @return the list of current option settings as an array of strings
   */
  public String[] getOptions(){

    String[] options = new String[5];
    int current = 0;

    options[current++] = "-G"; options[current++] = "" + m_NumAttemptsOfGene;
    options[current++] = "-I"; options[current++] = "" + m_NumFoldersMI;

    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 numAttemptsOfGeneOptionTipText() {
    return "Sets the number of attempts for generalization.";
  }

  /**
   * Gets the number of attempts for generalisation.
   *
   * @return the value of the option G
   */
  public int getNumAttemptsOfGeneOption() {
    return m_NumAttemptsOfGene;
  }


  /**
   * Sets the number of attempts for generalisation.
   *
   * @param newIntParameter the new value.
   */
  public void setNumAttemptsOfGeneOption(int newIntParameter) {
    m_NumAttemptsOfGene = newIntParameter;
  }

  /**
   * Returns the tip text for this property
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String numFoldersMIOptionTipText() {
    return "Sets the number of folder for mutual information.";
  }

  /**
   * Gets the number of folder for mutual information.
   *
   * @return the value of the option I
   */
  public int getNumFoldersMIOption() {
    return m_NumFoldersMI;
  }


  /**
   * Sets the number of folder for mutual information.
   *
   * @param newIntParameter the new value.
   */
  public void setNumFoldersMIOption(int newIntParameter) {
    m_NumFoldersMI = newIntParameter;
  }



  /** ENTRY POINT */


  /**
   * Main method for testing this class.
   *
   * @param argv should contain command line arguments for evaluation
   * (see Evaluation).
   */
  public static void main(String [] argv) {
    runClassifier(new NNge(), argv);
  }
}