clustererpanel.java

wekaUT是 university texas austin 开发的基于weka的半指导学习(semi supervised learning)的分类器

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.EAST;
    gbC.fill = GridBagConstraints.HORIZONTAL;
    gbC.gridy = 1;     gbC.gridx = 1;    gbC.gridwidth = 2;
    gbC.insets = new Insets(2, 10, 2, 0);
    gbL.setConstraints(m_SetTestBut, gbC);
    p2.add(m_SetTestBut);

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.WEST;
    gbC.gridy = 2;     gbC.gridx = 0;
    gbL.setConstraints(m_PercentBut, gbC);
    p2.add(m_PercentBut);

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.EAST;
    gbC.fill = GridBagConstraints.HORIZONTAL;
    gbC.gridy = 2;     gbC.gridx = 1;
    gbC.insets = new Insets(2, 10, 2, 10);
    gbL.setConstraints(m_PercentLab, gbC);
    p2.add(m_PercentLab);

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.EAST;
    gbC.fill = GridBagConstraints.HORIZONTAL;
    gbC.gridy = 2;     gbC.gridx = 2;  gbC.weightx = 100;
    gbC.ipadx = 20;
    gbL.setConstraints(m_PercentText, gbC);
    p2.add(m_PercentText);

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.WEST;
    gbC.gridy = 3;     gbC.gridx = 0;  gbC.gridwidth = 2;
    gbL.setConstraints(m_ClassesToClustersBut, gbC);
    p2.add(m_ClassesToClustersBut);

    m_ClassCombo.setBorder(BorderFactory.createEmptyBorder(0, 20, 0, 0));
    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.WEST;
    gbC.gridy = 4;     gbC.gridx = 0;  gbC.gridwidth = 2;
    gbL.setConstraints(m_ClassCombo, gbC);
    p2.add(m_ClassCombo);

    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.WEST;
    gbC.gridy = 5;     gbC.gridx = 0;  gbC.gridwidth = 2;
    gbL.setConstraints(m_StorePredictionsBut, gbC);
    p2.add(m_StorePredictionsBut);

    JPanel buttons = new JPanel();
    buttons.setLayout(new GridLayout(2, 1));
    JPanel ssButs = new JPanel();
    ssButs.setBorder(BorderFactory.createEmptyBorder(5, 5, 5, 5));
    ssButs.setLayout(new GridLayout(1, 2, 5, 5));
    ssButs.add(m_StartBut);
    ssButs.add(m_StopBut);

    JPanel ib = new JPanel();
    ib.setBorder(BorderFactory.createEmptyBorder(5, 5, 5, 5));
    ib.setLayout(new GridLayout(1, 1, 5, 5));
    ib.add(m_ignoreBut);
    buttons.add(ib);
    buttons.add(ssButs);
    
    JPanel p3 = new JPanel();
    p3.setBorder(BorderFactory.createTitledBorder("Clusterer output"));
    p3.setLayout(new BorderLayout());
    final JScrollPane js = new JScrollPane(m_OutText);
    p3.add(js, BorderLayout.CENTER);
    js.getViewport().addChangeListener(new ChangeListener() {
      private int lastHeight;
      public void stateChanged(ChangeEvent e) {
	JViewport vp = (JViewport)e.getSource();
	int h = vp.getViewSize().height; 
	if (h != lastHeight) { // i.e. an addition not just a user scrolling
	  lastHeight = h;
	  int x = h - vp.getExtentSize().height;
	  vp.setViewPosition(new Point(0, x));
	}
      }
    });    

    JPanel mondo = new JPanel();
    gbL = new GridBagLayout();
    mondo.setLayout(gbL);
    gbC = new GridBagConstraints();
    //    gbC.anchor = GridBagConstraints.WEST;
    gbC.fill = GridBagConstraints.HORIZONTAL;
    gbC.gridy = 0;     gbC.gridx = 0;
    gbL.setConstraints(p2, gbC);
    mondo.add(p2);
    gbC = new GridBagConstraints();
    gbC.anchor = GridBagConstraints.NORTH;
    gbC.fill = GridBagConstraints.HORIZONTAL;
    gbC.gridy = 1;     gbC.gridx = 0;
    gbL.setConstraints(buttons, gbC);
    mondo.add(buttons);
    gbC = new GridBagConstraints();
    //gbC.anchor = GridBagConstraints.NORTH;
    gbC.fill = GridBagConstraints.BOTH;
    gbC.gridy = 2;     gbC.gridx = 0; gbC.weightx = 0;
    gbL.setConstraints(m_History, gbC);
    mondo.add(m_History);
    gbC = new GridBagConstraints();
    gbC.fill = GridBagConstraints.BOTH;
    gbC.gridy = 0;     gbC.gridx = 1;
    gbC.gridheight = 3;
    gbC.weightx = 100; gbC.weighty = 100;
    gbL.setConstraints(p3, gbC);
    mondo.add(p3);

    setLayout(new BorderLayout());
    add(p1, BorderLayout.NORTH);
    add(mondo, BorderLayout.CENTER);
  }
  
  /**
   * Updates the enabled status of the input fields and labels.
   */
  protected void updateRadioLinks() {
    
    m_SetTestBut.setEnabled(m_TestSplitBut.isSelected());
    if ((m_SetTestFrame != null) && (!m_TestSplitBut.isSelected())) {
      m_SetTestFrame.setVisible(false);
    }
    m_PercentText.setEnabled(m_PercentBut.isSelected());
    m_PercentLab.setEnabled(m_PercentBut.isSelected());
    m_ClassCombo.setEnabled(m_ClassesToClustersBut.isSelected());
  }

  /**
   * Sets the Logger to receive informational messages
   *
   * @param newLog the Logger that will now get info messages
   */
  public void setLog(Logger newLog) {

    m_Log = newLog;
  }
  
  /**
   * Set the default attributes to use on the x and y axis
   * of a new visualization object.
   * @param x the index of the attribute to use on the x axis
   * @param y the index of the attribute to use on the y axis
   */
  public void setXY_VisualizeIndexes(int x, int y) {
    m_visXIndex = x;
    m_visYIndex = y;
  }

  /**
   * Tells the panel to use a new set of instances.
   *
   * @param inst a set of Instances
   */
  public void setInstances(Instances inst) {
    
    setXY_VisualizeIndexes(0,0);

    m_Instances = inst;
   
    m_ignoreKeyModel.removeAllElements();
    
    String [] attribNames = new String [m_Instances.numAttributes()];
    for (int i = 0; i < m_Instances.numAttributes(); i++) {
      String name = m_Instances.attribute(i).name();
      m_ignoreKeyModel.addElement(name);

       String type = "";
      switch (m_Instances.attribute(i).type()) {
      case Attribute.NOMINAL:
	type = "(Nom) ";
	break;
      case Attribute.NUMERIC:
	type = "(Num) ";
	break;
      case Attribute.STRING:
	type = "(Str) ";
	break;
      default:
	type = "(???) ";
      }
      String attnm = m_Instances.attribute(i).name();
     
      attribNames[i] = type + attnm;
    }

    
    m_StartBut.setEnabled(m_RunThread == null);
    m_StopBut.setEnabled(m_RunThread != null);
    m_ignoreBut.setEnabled(true);
    m_ClassCombo.setModel(new DefaultComboBoxModel(attribNames));
    m_ClassCombo.setSelectedIndex(attribNames.length - 1);
    updateRadioLinks();
  }

  /**
   * Sets the user test set. Information about the current test set
   * is displayed in an InstanceSummaryPanel and the user is given the
   * ability to load another set from a file or url.
   *
   */
  protected void setTestSet() {

    if (m_SetTestFrame == null) {
      final SetInstancesPanel sp = new SetInstancesPanel();
      m_Summary = sp.getSummary();
      if (m_TestInstancesCopy != null) {
	sp.setInstances(m_TestInstances);
      }
      sp.addPropertyChangeListener(new PropertyChangeListener() {
	public void propertyChange(PropertyChangeEvent e) {
	  m_TestInstances = sp.getInstances();
	}
      });
      // Add propertychangelistener to update m_TestInstances whenever
      // it changes in the settestframe
      m_SetTestFrame = new JFrame("Test Instances");
      m_SetTestFrame.getContentPane().setLayout(new BorderLayout());
      m_SetTestFrame.getContentPane().add(sp, BorderLayout.CENTER);
      m_SetTestFrame.pack();
    }
    m_SetTestFrame.setVisible(true);
  }

  /**
   * Sets up the structure for the visualizable instances. This dataset
   * contains the original attributes plus the clusterer's cluster assignments
   * @param testInstancs the instances that the clusterer has clustered
   * @return a PlotData2D object encapsulating the visualizable instances. The    * instances contain one more attribute (predicted
   * cluster) than the testInstances
   */
  private PlotData2D setUpVisualizableInstances(Instances testInstances,
						ClusterEvaluation eval) 
    throws Exception {

    int numClusters = eval.getNumClusters();
    double [] clusterAssignments = eval.getClusterAssignments();

    FastVector hv = new FastVector();
    Instances newInsts;

    Attribute predictedCluster;
    FastVector clustVals = new FastVector();

    for (int i = 0; i < numClusters; i++) {
      clustVals.addElement("cluster"+i);
    }
    predictedCluster = new Attribute("Cluster", clustVals);
    for (int i = 0; i < testInstances.numAttributes(); i++) {
      hv.addElement(testInstances.attribute(i).copy());
    }
    hv.addElement(predictedCluster);
    
    newInsts = new Instances(testInstances.relationName()+"_clustered", hv, 
			     testInstances.numInstances());

    double [] values;
    int j;
    int [] pointShapes = null;
    int [] classAssignments = null;
    if (testInstances.classIndex() >= 0) {
      classAssignments = eval.getClassesToClusters();
      pointShapes = new int[testInstances.numInstances()];
      for (int i = 0; i < testInstances.numInstances(); i++) {
	pointShapes[i] = Plot2D.CONST_AUTOMATIC_SHAPE;
      }
    }

    for (int i = 0; i < testInstances.numInstances(); i++) {
      values = new double[newInsts.numAttributes()];
      for (j = 0; j < testInstances.numAttributes(); j++) {
	values[j] = testInstances.instance(i).value(j);
      }
      values[j] = clusterAssignments[i];
      newInsts.add(new Instance(1.0, values));
      if (pointShapes != null) {
	if ((int)testInstances.instance(i).classValue() != 
	    classAssignments[(int)clusterAssignments[i]]) {
	  pointShapes[i] = Plot2D.ERROR_SHAPE;
	}
      }
    }
    PlotData2D plotData = new PlotData2D(newInsts);
    if (pointShapes != null) {
      plotData.setShapeType(pointShapes);
    }
    plotData.addInstanceNumberAttribute();
    return plotData;
  }
  
  /**
   * Starts running the currently configured clusterer with the current
   * settings. This is run in a separate thread, and will only start if
   * there is no clusterer already running. The clusterer output is sent
   * to the results history panel.
   */
  protected void startClusterer() {

    if (m_RunThread == null) {
      m_StartBut.setEnabled(false);
      m_StopBut.setEnabled(true);
      m_ignoreBut.setEnabled(false);
      m_RunThread = new Thread() {
	public void run() {
	  // Copy the current state of things
	  m_Log.statusMessage("Setting up...");
	  Instances inst = new Instances(m_Instances);
	  Instances userTest = null;
	  PlotData2D predData = null;
	  if (m_TestInstances != null) {
	    userTest = new Instances(m_TestInstancesCopy);
	  }
	  
	  boolean saveVis = m_StorePredictionsBut.isSelected();
	  String grph = null;
	  int[] ignoredAtts = null;

	  int testMode = 0;
	  int percent = 66;
	  Clusterer clusterer = (Clusterer) m_ClustererEditor.getValue();
	  StringBuffer outBuff = new StringBuffer();
	  String name = (new SimpleDateFormat("HH:mm:ss - "))
	  .format(new Date());
	  String cname = clusterer.getClass().getName();
	  if (cname.startsWith("weka.clusterers.")) {
	    name += cname.substring("weka.clusterers.".length());
	  } else {
	    name += cname;
	  }
	  try {
	    m_Log.logMessage("Started " + cname);
	    if (m_Log instanceof TaskLogger) {
	      ((TaskLogger)m_Log).taskStarted();
	    }
	    if (m_PercentBut.isSelected()) {
	      testMode = 2;
	      percent = Integer.parseInt(m_PercentText.getText());
	      if ((percent <= 0) || (percent >= 100)) {
		throw new Exception("Percentage must be between 0 and 100");
	      }
	    } else if (m_TrainBut.isSelected()) {
	      testMode = 3;
	    } else if (m_TestSplitBut.isSelected()) {
	      testMode = 4;
	      // Check the test instance compatibility
	      if (userTest == null) {
		throw new Exception("No user test set has been opened");
	      }
	      if (!inst.equalHeaders(userTest)) {
		throw new Exception("Train and test set are not compatible");
	      }
	    } else if (m_ClassesToClustersBut.isSelected()) {
	      testMode = 5;
	    } else {
	      throw new Exception("Unknown test mode");
	    }

	    Instances trainInst = new Instances(inst);
	    if (m_ClassesToClustersBut.isSelected()) {
	      trainInst.setClassIndex(m_ClassCombo.getSelectedIndex());
	      inst.setClassIndex(m_ClassCombo.getSelectedIndex());
	      if (inst.classAttribute().isNumeric()) {
		throw new Exception("Class must be nominal for class based "
				    +"evaluation!");
	      }
	    }
	    if (!m_ignoreKeyList.isSelectionEmpty()) {
	      trainInst = removeIgnoreCols(trainInst);
	    }

	    // Output some header information
	    outBuff.append("=== Run information ===\n\n");
	    outBuff.append("Scheme:       " + cname);
	    if (clusterer instanceof OptionHandler) {
	      String [] o = ((OptionHandler) clusterer).getOptions();
	      outBuff.append(" " + Utils.joinOptions(o));
	    }
	    outBuff.append("\n");
	    outBuff.append("Relation:     " + inst.relationName() + '\n');
	    outBuff.append("Instances:    " + inst.numInstances() + '\n');
	    outBuff.append("Attributes:   " + inst.numAttributes() + '\n');
	    if (inst.numAttributes() < 100) {
	      boolean [] selected = new boolean [inst.numAttributes()];
	      for (int i = 0; i < inst.numAttributes(); i++) {
		selected[i] = true;
	      }
	      if (!m_ignoreKeyList.isSelectionEmpty()) {
		int [] indices = m_ignoreKeyList.getSelectedIndices();
		for (int i = 0; i < indices.length; i++) {