clustererpanel.java

:<<数据挖掘--实用机器学习技术及java实现>>一书的配套源程序

		}
	      }
	      if (m_ClassesToClustersBut.isSelected()) {
		selected[m_ClassCombo.getSelectedIndex()] = false;
	      }
	      for (int i = 0; i < inst.numAttributes(); i++) {
		if (selected[i]) {
		  outBuff.append("              " + inst.attribute(i).name()
				 + '\n');
		}
	      }
	      if (!m_ignoreKeyList.isSelectionEmpty() 
		  || m_ClassesToClustersBut.isSelected()) {
		outBuff.append("Ingored:\n");
		for (int i = 0; i < inst.numAttributes(); i++) {
		  if (!selected[i]) {
		    outBuff.append("              " + inst.attribute(i).name()
				   + '\n');
		  }
		}
	      }
	    } else {
	      outBuff.append("              [list of attributes omitted]\n");
	    }

	    outBuff.append("Test mode:    ");
	    switch (testMode) {
	      case 3: // Test on training
	      outBuff.append("evaluate on training data\n");
	      break;
	      case 2: // Percent split
	      outBuff.append("split " + percent
			       + "% train, remainder test\n");
	      break;
	      case 4: // Test on user split
	      outBuff.append("user supplied test set: "
			     + userTest.numInstances() + " instances\n");
	      break;
	    case 5: // Classes to clusters evaluation on training
	      outBuff.append("Classes to clusters evaluation on training data");
	      
	      break;
	    }
	    outBuff.append("\n");
	    m_History.addResult(name, outBuff);
	    m_History.setSingle(name);
	    
	    // Build the model and output it.
	    m_Log.statusMessage("Building model on training data...");

	    // remove the class attribute (if set) and build the clusterer
	    clusterer.buildClusterer(removeClass(trainInst));

	    outBuff.append("\n=== Clustering model (full training set) ===\n\n");
	    outBuff.append(clusterer.toString() + '\n');
	    m_History.updateResult(name);
	    
	    ClusterEvaluation eval = new ClusterEvaluation();
	    eval.setClusterer(clusterer);
	    switch (testMode) {
	      case 3: case 5: // Test on training
	      m_Log.statusMessage("Clustering training data...");
	      eval.evaluateClusterer(trainInst);
	      predData = setUpVisualizableInstances(inst,eval);
	      outBuff.append("=== Evaluation on training set ===\n\n");
	      break;

	      case 2: // Percent split
	      m_Log.statusMessage("Randomizing instances...");
	      inst.randomize(new Random(42));
	      trainInst.randomize(new Random(42));
	      int trainSize = trainInst.numInstances() * percent / 100;
	      int testSize = trainInst.numInstances() - trainSize;
	      Instances train = new Instances(trainInst, 0, trainSize);
	      Instances test = new Instances(trainInst, trainSize, testSize);
	      Instances testVis = new Instances(inst, trainSize, testSize);
	      m_Log.statusMessage("Building model on training split...");
	      clusterer.buildClusterer(train);
	      m_Log.statusMessage("Evaluating on test split...");
	      eval.evaluateClusterer(test);
	      predData = setUpVisualizableInstances(testVis, eval);
	      outBuff.append("=== Evaluation on test split ===\n");
	      break;
		
	      case 4: // Test on user split
	      m_Log.statusMessage("Evaluating on test data...");
	      Instances userTestT = new Instances(userTest);
	      if (!m_ignoreKeyList.isSelectionEmpty()) {
		userTestT = removeIgnoreCols(userTestT);
	      }
	      eval.evaluateClusterer(userTestT);
	      predData = setUpVisualizableInstances(userTest, eval);
	      outBuff.append("=== Evaluation on test set ===\n");
	      break;

	      default:
	      throw new Exception("Test mode not implemented");
	    }
	    outBuff.append(eval.clusterResultsToString());
	    outBuff.append("\n");
	    m_History.updateResult(name);
	    m_Log.logMessage("Finished " + cname);
	    m_Log.statusMessage("OK");
	  } catch (Exception ex) {
	    m_Log.logMessage(ex.getMessage());
	    m_Log.statusMessage("See error log");
	    ex.printStackTrace();
	  } finally {
	    if (predData != null) {
	      m_CurrentVis = new VisualizePanel();
	      m_CurrentVis.setName(name+" ("+inst.relationName()+")");
	      m_CurrentVis.setLog(m_Log);
	      predData.setPlotName(name+" ("+inst.relationName()+")");
	      
	      try {
		m_CurrentVis.addPlot(predData);
		m_CurrentVis.setXIndex(m_visXIndex); 
		m_CurrentVis.setYIndex(m_visYIndex);
	      } catch (Exception ex) {
		System.err.println(ex);
	      }
	      m_CurrentVis.addActionListener(new ActionListener() {
		  public void actionPerformed(ActionEvent e) {
		    if (m_CurrentVis.getInstances().
			relationName().
			compareTo(m_Instances.relationName()) == 0) {
		      setXY_VisualizeIndexes(m_CurrentVis.getXIndex(), 
					     m_CurrentVis.getYIndex());
		    }
		  }
		});

	      if (saveVis) {
		FastVector vv = new FastVector();
		vv.addElement(m_CurrentVis);
		m_History.addObject(name, vv);
	      }
	    }
	    if (isInterrupted()) {
	      m_Log.logMessage("Interrupted " + cname);
	      m_Log.statusMessage("See error log");
	    }
	    m_RunThread = null;
	    m_StartBut.setEnabled(true);
	    m_StopBut.setEnabled(false);
	    m_ignoreBut.setEnabled(true);
	    if (m_Log instanceof TaskLogger) {
	      ((TaskLogger)m_Log).taskFinished();
	    }
	  }
	}
      };
      m_RunThread.setPriority(Thread.MIN_PRIORITY);
      m_RunThread.start();
    }
  }

  private Instances removeClass(Instances inst) {
    AttributeFilter af = new AttributeFilter();
    Instances retI = null;
    
    try {
      if (inst.classIndex() < 0) {
	retI = inst;
      } else {
	af.setAttributeIndices(""+(inst.classIndex()+1));
	af.setInvertSelection(false);
	af.setInputFormat(inst);
	retI = Filter.useFilter(inst, af);
      }
    } catch (Exception e) {
      e.printStackTrace();
    }
    return retI;
  }

  private Instances removeIgnoreCols(Instances inst) {
    
    // If the user is doing classes to clusters evaluation and
    // they have opted to ignore the class, then unselect the class in
    // the ignore list
    if (m_ClassesToClustersBut.isSelected()) {
      int classIndex = m_ClassCombo.getSelectedIndex();
      if (m_ignoreKeyList.isSelectedIndex(classIndex)) {
	m_ignoreKeyList.removeSelectionInterval(classIndex, classIndex);
      }
    }
    int [] selected = m_ignoreKeyList.getSelectedIndices();
    AttributeFilter af = new AttributeFilter();
    Instances retI = null;

    try {
      af.setAttributeIndicesArray(selected);
      af.setInvertSelection(false);
      af.setInputFormat(inst);
      retI = Filter.useFilter(inst, af);
    } catch (Exception e) {
      e.printStackTrace();
    }
    
    return retI;
  }

  /**
   * Stops the currently running clusterer (if any).
   */
  protected void stopClusterer() {

    if (m_RunThread != null) {
      m_RunThread.interrupt();
      
      // This is deprecated (and theoretically the interrupt should do).
      m_RunThread.stop();
      
    }
  }

  /**
   * Pops up a TreeVisualizer for the clusterer from the currently
   * selected item in the results list
   * @param dottyString the description of the tree in dotty format
   * @param treeName the title to assign to the display
   */
  protected void visualizeTree(String dottyString, String treeName) {
    final javax.swing.JFrame jf = 
      new javax.swing.JFrame("Weka Classifier Tree Visualizer: "+treeName);
    jf.setSize(500,400);
    jf.getContentPane().setLayout(new BorderLayout());
    TreeVisualizer tv = new TreeVisualizer(null,
					   dottyString,
					   new PlaceNode2());
    jf.getContentPane().add(tv, BorderLayout.CENTER);
    jf.addWindowListener(new java.awt.event.WindowAdapter() {
	public void windowClosing(java.awt.event.WindowEvent e) {
	  jf.dispose();
	}
      });
    
    jf.setVisible(true);
  }

  /**
   * Pops up a visualize panel to display cluster assignments
   * @param sp the visualize panel to display
   */
  protected void visualizeClusterAssignments(VisualizePanel sp) {
    if (sp != null) {
      String plotName = sp.getName();
      final javax.swing.JFrame jf = 
	new javax.swing.JFrame("Weka Clusterer Visualize: "+plotName);
      jf.setSize(500,400);
      jf.getContentPane().setLayout(new BorderLayout());
      jf.getContentPane().add(sp, BorderLayout.CENTER);
      jf.addWindowListener(new java.awt.event.WindowAdapter() {
	  public void windowClosing(java.awt.event.WindowEvent e) {
	    jf.dispose();
	  }
	});

      jf.setVisible(true);
    }
  }

  /**
   * Handles constructing a popup menu with visualization options
   * @param name the name of the result history list entry clicked on by
   * the user
   * @param x the x coordinate for popping up the menu
   * @param y the y coordinate for popping up the menu
   */
  protected void visualizeClusterer(String name, int x, int y) {
    final String selectedName = name;
    JPopupMenu resultListMenu = new JPopupMenu();

    JMenuItem visMainBuffer = new JMenuItem("View in main window");
    visMainBuffer.addActionListener(new ActionListener() {
	public void actionPerformed(ActionEvent e) {
	  m_History.setSingle(selectedName);
	}
      });
    resultListMenu.add(visMainBuffer);

    JMenuItem visSepBuffer = new JMenuItem("View in separate window");
    visSepBuffer.addActionListener(new ActionListener() {
	public void actionPerformed(ActionEvent e) {
	  m_History.openFrame(selectedName);
	}
      });
    resultListMenu.add(visSepBuffer);

    JMenuItem saveOutput = new JMenuItem("Save result buffer");
    saveOutput.addActionListener(new ActionListener() {
	public void actionPerformed(ActionEvent e) {
	  saveBuffer(selectedName);
	}
      });
    resultListMenu.add(saveOutput);
    resultListMenu.addSeparator();

    FastVector o = (FastVector)m_History.getNamedObject(selectedName);

    if (o != null) {
      VisualizePanel temp_vp = null;
      String temp_grph = null;
     
      for (int i = 0; i < o.size(); i++) {
	Object temp = o.elementAt(i);
	if (temp instanceof VisualizePanel) { // normal errors
	  temp_vp = (VisualizePanel)temp;
	} else if (temp instanceof String) { // graphable output
	  temp_grph = (String)temp;
	}
      }

      final VisualizePanel vp = temp_vp;
      final String grph = temp_grph;

      JMenuItem visClusts = new JMenuItem("Visualize cluster assignments");
      if (vp != null) {
	visClusts.addActionListener(new ActionListener() {
	    public void actionPerformed(ActionEvent e) {
	      visualizeClusterAssignments(vp);
	    }
	  });
	resultListMenu.add(visClusts);
      }
      JMenuItem visTree = new JMenuItem("Visualize tree");
      if (grph != null) {
	visTree.addActionListener(new ActionListener() {
	    public void actionPerformed(ActionEvent e) {
	      visualizeTree(grph,vp.getName());
	    }
	  });
	resultListMenu.add(visTree);
      }
      
    }
    resultListMenu.show(m_History.getList(), x, y);
  }

  /**
   * Save the currently selected clusterer output to a file.
   * @param name the name of the buffer to save
   */
  protected void saveBuffer(String name) {
    StringBuffer sb = m_History.getNamedBuffer(name);
    if (sb != null) {
      if (m_SaveOut.save(sb)) {
	m_Log.logMessage("Save succesful.");
      }
    }
  }

  private void setIgnoreColumns() {
    ListSelectorDialog jd = new ListSelectorDialog(null, m_ignoreKeyList);

    // Open the dialog
    int result = jd.showDialog();
    
    if (result != ListSelectorDialog.APPROVE_OPTION) {
      // clear selected indices
      m_ignoreKeyList.clearSelection();
    }
  }

  /**
   * Tests out the clusterer panel from the command line.
   *
   * @param args may optionally contain the name of a dataset to load.
   */
  public static void main(String [] args) {

    try {
      final javax.swing.JFrame jf =
	new javax.swing.JFrame("Weka Knowledge Explorer: Cluster");
      jf.getContentPane().setLayout(new BorderLayout());
      final ClustererPanel sp = new ClustererPanel();
      jf.getContentPane().add(sp, BorderLayout.CENTER);
      weka.gui.LogPanel lp = new weka.gui.LogPanel();
      sp.setLog(lp);
      jf.getContentPane().add(lp, BorderLayout.SOUTH);
      jf.addWindowListener(new java.awt.event.WindowAdapter() {
	public void windowClosing(java.awt.event.WindowEvent e) {
	  jf.dispose();
	  System.exit(0);
	}
      });
      jf.pack();
      jf.setSize(800, 600);
      jf.setVisible(true);
      if (args.length == 1) {
	System.err.println("Loading instances from " + args[0]);
	java.io.Reader r = new java.io.BufferedReader(
			   new java.io.FileReader(args[0]));
	Instances i = new Instances(r);
	sp.setInstances(i);
      }
    } catch (Exception ex) {
      ex.printStackTrace();
      System.err.println(ex.getMessage());
    }
  }
}