userclassifier.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

    return result;
  }

  /**
   * Returns default capabilities of the classifier.
   *
   * @return      the capabilities of this classifier
   */
  public Capabilities getCapabilities() {
    Capabilities result = super.getCapabilities();

    // attributes
    result.enable(Capability.NOMINAL_ATTRIBUTES);
    result.enable(Capability.NUMERIC_ATTRIBUTES);
    result.enable(Capability.DATE_ATTRIBUTES);
    result.enable(Capability.STRING_ATTRIBUTES);
    result.enable(Capability.RELATIONAL_ATTRIBUTES);
    result.enable(Capability.MISSING_VALUES);

    // class
    result.enable(Capability.NOMINAL_CLASS);
    result.enable(Capability.NUMERIC_CLASS);
    result.enable(Capability.DATE_CLASS);
    result.enable(Capability.MISSING_CLASS_VALUES);

    // instances
    result.setMinimumNumberInstances(0);
    
    return result;
  }

  /**
   * Call this function to build a decision tree for the training
   * data provided.
   * @param i The training data.
   * @throws Exception if can't build classification properly.
   */
  public void buildClassifier(Instances i) throws Exception {
    // can classifier handle the data?
    getCapabilities().testWithFail(i);

    // remove instances with missing class
    i = new Instances(i);
    i.deleteWithMissingClass();
    
    //construct a visualizer
    //construct a tree displayer and feed both then do nothing
    //note that I will display at the bottom of each split how many 
    //fall into each catagory
    
    m_classifiers = new GenericObjectEditor(true);
    m_classifiers.setClassType(Classifier.class);
    m_classifiers.setValue(new weka.classifiers.rules.ZeroR());
    
    ((GenericObjectEditor.GOEPanel)m_classifiers.getCustomEditor())
      .addOkListener(new ActionListener() {
	  public void actionPerformed(ActionEvent e) {
	    //I want to use the focus variable but to trust it I need
	    //to kill the window if anything gets changed by either
	    //editor
	    try {
	      m_focus.m_set1 = null;
	      m_focus.m_set2 = null;
	      m_focus.setInfo(m_focus.m_attrib1, m_focus.m_attrib2, null);
	      m_focus.setClassifier((Classifier)m_classifiers.getValue());
	      m_classifiers = new GenericObjectEditor();
	      m_classifiers.setClassType(Classifier.class);
	      m_classifiers.setValue(new weka.classifiers.rules.ZeroR());
	      ((GenericObjectEditor.GOEPanel)m_classifiers.getCustomEditor())
		.addOkListener(this);
	      m_tView = new TreeVisualizer(UserClassifier.this, graph(), 
				     new PlaceNode2());
	      m_tView.setHighlight(m_focus.m_identity);
	      m_reps.setComponentAt(0, m_tView);
	      m_iView.setShapes(null);
	    } catch(Exception er) {
	      System.out.println("Error : " + er);
	      System.out.println("Part of user input so had to catch here");
	    }
	  }
	});
    
    m_built = false;
    m_mainWin = new JFrame();
    
    m_mainWin.addWindowListener(new WindowAdapter() {
	public void windowClosing(WindowEvent e) {
	  int well = JOptionPane.showConfirmDialog(m_mainWin, 
						   "Are You Sure...\n"
						   + "Click Yes To Accept"
						   + " The Tree" 
						   + "\n Click No To Return",
						   "Accept Tree", 
						   JOptionPane.YES_NO_OPTION);
	  
	  if (well == 0) {
	    m_mainWin.setDefaultCloseOperation(JFrame.DISPOSE_ON_CLOSE);
	    blocker(false);
	    
	  }
	  else {
	    m_mainWin.setDefaultCloseOperation(JFrame.DO_NOTHING_ON_CLOSE);
	  }
	}
      });
    
    m_reps = new JTabbedPane();
    m_mainWin.getContentPane().add(m_reps);
    
    //make a backup of the instances so that any changes don't go past here.
    Instances te = new Instances(i, i.numInstances());
    for (int noa = 0; noa < i.numInstances(); noa++) {
      te.add(i.instance(noa));
    }
    
    te.deleteWithMissingClass(); //remove all instances with a missing class
    //from training
    
    m_top = new TreeClass(null, 0, 0, m_nextId, 1, te, null);
    m_focus = m_top;
    //System.out.println(graph());
    m_tView = new TreeVisualizer(this, graph(), new PlaceNode1());
    
    m_reps.add("Tree Visualizer", m_tView);
    //tree_frame = new JFrame();
    //tree_frame.getContentPane().add(m_tView);
    //tree_frame.setSize(800,600);
    //tree_frame.setVisible(true);
    
    m_tView.setHighlight(m_top.m_identity);
    m_iView = new VisualizePanel(this);
    //m_iView.setSize(400, 300);
    m_iView.setInstances(m_top.m_training);
    m_iView.setColourIndex(te.classIndex());
    //vis_frame = new JFrame();
    //vis_frame.getContentPane().add(m_iView);
    //vis_frame.setSize(400, 300);
    //vis_frame.setVisible(true);
    m_reps.add("Data Visualizer", m_iView);
    m_mainWin.setSize(560, 420);
    m_mainWin.setVisible(true);
    blocker(true);          //a call so that the main thread of 
    //execution has to wait for the all clear message from the user.
    
    //so that it can be garbage 
    if (m_propertyDialog != null) {
      m_propertyDialog.dispose();
      m_propertyDialog = null;
    }
    
    //collected
    m_classifiers = null;
    m_built = true;
  }

  /**
   * Call this function to get a double array filled with the probability
   * of how likely each class type is the class of the instance.
   * @param i The instance to classify.
   * @return A double array filled with the probalities of each class type.
   * @throws Exception if can't classify instance.
   */
  public double[] distributionForInstance(Instance i) throws Exception {

    if (!m_built) {
      return null;
    }
    
    double[] res = m_top.calcClassType(i);
    if (m_top.m_training.classAttribute().isNumeric()) {
      return res;
    }

    double most_likely = 0, highest = -1;
    double count = 0;
    for (int noa = 0; noa < m_top.m_training.numClasses(); noa++) {
      count += res[noa];
      if (res[noa] > highest) {
	most_likely = noa;
	highest = res[noa];
      }
    }
    
    if (count <= 0) {
      //not sure how this happened.
      return null;
    }

    for (int noa = 0; noa < m_top.m_training.numClasses(); noa++) {
      res[noa] = res[noa] / count;
    }
    //System.out.println("ret");
    
    return res;
  }
  
  /**
   * Inner class used to represent the actual decision tree structure and data.
   */
  private class TreeClass 
    implements Serializable {
    
    /** for serialization */
    static final long serialVersionUID = 595663560871347434L;
    
    /**
     * This contains the info for the coords of the shape converted 
     * to attrib coords, 
     * for polygon the first attrib is the number of points, 
     * This is not more object oriented because that would 
     * be over kill.
     */
    public FastVector m_ranges;

    /** the first attribute */
    public int m_attrib1;
    
    /** the second attribute */
    public int m_attrib2;
    
    public TreeClass m_set1;
    public TreeClass m_set2;

    /** the parent */
    public TreeClass m_parent;

    /** A string to uniquely identify this node. */
    public String m_identity;
    
    /** the weight of this node */
    public double m_weight;
    
    /** the training instances for this node */
    public Instances m_training;
    
    /** Used instead of the standard leaf if one exists. */
    public Classifier m_classObject;

    /** Used on the instances while classifying if one exists. */
    public Filter m_filter;
    
    /**
     * Constructs a TreeClass node  with all the important information.
     * @param r A FastVector containing the shapes, null if it's a leaf node.
     * @param a1 The first attribute.
     * @param a2 The second attribute.
     * @param id The unique id number for this node.
     * @param w The weight of this node.
     * @param i The instances that make it to this node from the training data.
     * @param p the parent
     * @throws Exception if can't use 'i' properly.
     */
    public TreeClass(FastVector r, int a1, int a2, int id, double w, 
		     Instances i, TreeClass p) throws Exception {
      m_set1 = null;
      m_set2 = null;
      m_ranges = r;
      m_classObject = null;
      m_filter = null;
      m_training = i;
      m_attrib1 = a1;
      m_attrib2 = a2;
      m_identity = "N" + String.valueOf(id);
      m_weight = w;
      m_parent = p;
      m_nextId++;
      if (m_ranges == null) {
	
	setLeaf();
	//this will fill the ranges array with the 
	//number of times each class type occurs for the instances.
	/*m_ranges = new FastVector(1);
	  m_ranges.addElement(new FastVector(i.numClasses() + 1));
	  FastVector tmp = (FastVector)m_ranges.elementAt(0);
	  tmp.addElement(new Double(0));
	  for (int noa = 0; noa < i.numClasses(); noa++) {
	  tmp.addElement(new Double(0));
	  }
	  for (int noa = 0; noa < i.numInstances(); noa++) {
	  tmp.setElementAt(new Double(((Double)tmp.elementAt
	  ((int)i.instance(noa).
	  classValue() + 1)).doubleValue() + 
	  i.instance(noa).weight()),
	  (int)i.instance(noa).classValue() + 1);  
	  //this gets the current class value and alters it and replaces it
	  }*/
      }     
    }
    
    /**
     * Call this to set an alternate classifier For this node.
     * @param c The alternative classifier to use.
     * @throws Exception if alternate classifier can't build classification.
     */
    public void setClassifier(Classifier c) throws Exception {
      m_classObject = c;
      m_classObject.buildClassifier(m_training);
    }
    
    /**
     * Call this to set this node with different information to what
     * it was created with.
     * @param at1 The first attribute.
     * @param at2 The second attribute.
     * @param ar The shapes at this node, null if leaf node, or 
     * alternate classifier.
     * @throws Exception if leaf node and cant't create leaf info.
     */
    public void setInfo(int at1, int at2, FastVector ar) throws Exception {
      m_classObject = null;
      m_filter = null;
      m_attrib1 = at1;
      m_attrib2 = at2;
      m_ranges = ar;
      
      //FastVector tmp;
      if (m_ranges == null) {
	setLeaf();
	/*
	//this will fill the ranges array with the number of times 
	//each class type occurs for the instances.
	  if (m_training != null) {
	    m_ranges = new FastVector(1);
	    m_ranges.addElement(new FastVector(m_training.numClasses() + 1));
	    tmp = (FastVector)m_ranges.elementAt(0);
	    tmp.addElement(new Double(0));
	    for (int noa = 0; noa < m_training.numClasses(); noa++) {
	      tmp.addElement(new Double(0));
	    }
	    for (int noa = 0; noa < m_training.numInstances(); noa++) {
	      tmp.setElementAt(new Double(((Double)tmp.elementAt
					   ((int)m_training.instance(noa).
					    classValue() + 1)).doubleValue() + 
					  m_training.instance(noa).weight()), 
			       (int)m_training.instance(noa).classValue() + 1);
	      //this gets the current class val and alters it and replaces it
	      }
	      }*/
      }
    }
    
    /**
     * This sets up the informtion about this node such as the s.d or the
     * number of each class.
     * @throws Exception if problem with training instances.
     */
    private void setLeaf() throws Exception {
      //this will fill the ranges array with the number of times 
      //each class type occurs for the instances.
      //System.out.println("ihere");
      if (m_training != null ) {
	
	if (m_training.classAttribute().isNominal()) {
	  FastVector tmp;
	  
	  //System.out.println("ehlpe");
	  m_ranges = new FastVector(1);
	  m_ranges.addElement(new FastVector(m_training.numClasses() + 1));
	  tmp = (FastVector)m_ranges.elementAt(0);
	  tmp.addElement(new Double(0));
	  for (int noa = 0; noa < m_training.numClasses(); noa++) {
	    tmp.addElement(new Double(0));
	  }
	  for (int noa = 0; noa < m_training.numInstances(); noa++) {
	    tmp.setElementAt(new Double(((Double)tmp.elementAt
					 ((int)m_training.instance(noa).
					  classValue() + 1)).doubleValue() + 
					m_training.instance(noa).weight()), 
			     (int)m_training.instance(noa).classValue() + 1);
	    //this gets the current class val and alters it and replaces it
	  }
	}
	else {
	  //then calc the standard deviation.
	  m_ranges = new FastVector(1);
	  double t1 = 0;
	  for (int noa = 0; noa < m_training.numInstances(); noa++) {
	    t1 += m_training.instance(noa).classValue();
	  }
	  
	  if (m_training.numInstances() != 0) {
	    t1 /= m_training.numInstances();
	  }
	  double t2 = 0;
	  for (int noa = 0; noa < m_training.numInstances(); noa++) {
	    t2 += Math.pow(m_training.instance(noa).classValue() - t1, 2);
	  }
	  FastVector tmp;
	  if (m_training.numInstances() != 0) {
	    t1 = Math.sqrt(t2 / m_training.numInstances());
	    m_ranges.addElement(new FastVector(2));
	    tmp = (FastVector)m_ranges.elementAt(0);
	    tmp.addElement(new Double(0));
	    tmp.addElement(new Double(t1));
	  }
	  else {
	    m_ranges.addElement(new FastVector(2));
	    tmp = (FastVector)m_ranges.elementAt(0);
	    tmp.addElement(new Double(0));
	    tmp.addElement(new Double(Double.NaN));
	  }
	}
      }
    }

    /**
     * This will recursively go through the tree and return inside the 
     * array the weightings of each of the class types
     * for this instance. Note that this function returns an otherwise 
     * unreferenced double array so there are no worry's about
     * making changes.
     *
     * @param i The instance to test
     * @return A double array containing the results.
     * @throws Exception if can't use instance i properly.
     */
    public double[] calcClassType(Instance i) throws Exception {
      //note that it will be the same calcs for both numeric and nominal
      //attrib types.
      //note the weightings for returning stuff will need to be modified 
      //to work properly but will do for now.
      double x = 0, y = 0;
      if (m_attrib1 >= 0) {
	x = i.value(m_attrib1);
      }
      if (m_attrib2 >= 0) {
	y = i.value(m_attrib2);
      }
      double[] rt;
      if (m_training.classAttribute().isNominal()) {
	rt = new double[m_training.numClasses()];
      }
      else {
	rt = new double[1];
      }

      FastVector tmp;
      if (m_classObject != null) {
	//then use the classifier.
	if (m_training.classAttribute().isNominal()) {
	  rt[(int)m_classObject.classifyInstance(i)] = 1;
	}
	else {
	  if (m_filter != null) {
	    m_filter.input(i);
	    rt[0] = m_classObject.classifyInstance(m_filter.output());
	  }
	  else {
	    rt[0] = m_classObject.classifyInstance(i);
	  }
	}
	//System.out.println("j48");
	return rt;
      }
      else if (((Double)((FastVector)m_ranges.elementAt(0)).
		elementAt(0)).intValue() == LEAF) {
	//System.out.println("leaf");
	//then this is a leaf
	//rt = new double[m_training.numClasses()];
	
	if (m_training.classAttribute().isNumeric()) {
	 
	  setLinear();
	  m_filter.input(i);
	  rt[0] = m_classObject.classifyInstance(m_filter.output());
	  return rt;
	}
	
	int totaler = 0;
	tmp = (FastVector)m_ranges.elementAt(0);
	for (int noa = 0; noa < m_training.numClasses();noa++) {
	  rt[noa] = ((Double)tmp.elementAt(noa + 1)).doubleValue();
	  totaler += rt[noa];
	}
	for (int noa = 0; noa < m_training.numClasses(); noa++) {
	  rt[noa] = rt[noa] / totaler;
	}
	return rt;
      }
      
      for (int noa = 0; noa < m_ranges.size(); noa++) {
	
	tmp = (FastVector)m_ranges.elementAt(noa);
	
	if (((Double)tmp.elementAt(0)).intValue() 
	    == VLINE && !Instance.isMissingValue(x)) {
	  
	}
	else if (((Double)tmp.elementAt(0)).intValue()