clustererpanel.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

		ignoredAtts[0] = m_ClassCombo.getSelectedIndex();
	      } else {
		int[] newIgnoredAtts = new int[ignoredAtts.length+1];
		System.arraycopy(ignoredAtts, 0, newIgnoredAtts, 0, ignoredAtts.length);
		newIgnoredAtts[ignoredAtts.length] = m_ClassCombo.getSelectedIndex();
		ignoredAtts = newIgnoredAtts;
	      }
	    }


	    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));
	    
	    if (testMode == 2) {
	      outBuff.append("\n=== Clustering model (full training set) ===\n\n");
	    
	      outBuff.append(clusterer.toString() + '\n');
	    }
	    m_History.updateResult(name);
	    if (clusterer instanceof Drawable) {
	      try {
		grph = ((Drawable)clusterer).graph();
	      } catch (Exception ex) {
	      }
	    }
	    // copy full model for output
	    SerializedObject so = new SerializedObject(clusterer);
	    fullClusterer = (Clusterer) so.getObject();
	    
	    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("=== Model and evaluation on training set ===\n\n");
	      break;

	      case 2: // Percent split
	      m_Log.statusMessage("Randomizing instances...");
	      inst.randomize(new Random(1));
	      trainInst.randomize(new Random(1));
	      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("=== Model and 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("=== Model and 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) {
	    ex.printStackTrace();
	    m_Log.logMessage(ex.getMessage());
	    JOptionPane.showMessageDialog(ClustererPanel.this,
					  "Problem evaluating clusterer:\n"
					  + ex.getMessage(),
					  "Evaluate clusterer",
					  JOptionPane.ERROR_MESSAGE);
	    m_Log.statusMessage("Problem evaluating clusterer");
	  } 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);
	      } catch (Exception ex) {
		System.err.println(ex);
	      }

	      FastVector vv = new FastVector();
	      vv.addElement(fullClusterer);
	      Instances trainHeader = new Instances(m_Instances, 0);
	      vv.addElement(trainHeader);
	      if (ignoredAtts != null) vv.addElement(ignoredAtts);
	      if (saveVis) {
		vv.addElement(m_CurrentVis);
		if (grph != null) {
		  vv.addElement(grph);
		}
		
	      }
	      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) {
    Remove af = new Remove();
    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();
    Remove af = new Remove();
    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;
  }

  private Instances removeIgnoreCols(Instances inst, int[] toIgnore) {

    Remove af = new Remove();
    Instances retI = null;

    try {
      af.setAttributeIndicesArray(toIgnore);
      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);
    tv.fitToScreen();
  }

  /**
   * 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");
    if (selectedName != null) {
      visMainBuffer.addActionListener(new ActionListener() {
	  public void actionPerformed(ActionEvent e) {
	    m_History.setSingle(selectedName);
	  }
	});
    } else {
      visMainBuffer.setEnabled(false);
    }
    resultListMenu.add(visMainBuffer);

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

    JMenuItem saveOutput = new JMenuItem("Save result buffer");
    if (selectedName != null) {
      saveOutput.addActionListener(new ActionListener() {
	  public void actionPerformed(ActionEvent e) {
	    saveBuffer(selectedName);
	  }
	});
    } else {
      saveOutput.setEnabled(false);
    }
    resultListMenu.add(saveOutput);
    
    JMenuItem deleteOutput = new JMenuItem("Delete result buffer");
    if (selectedName != null) {
      deleteOutput.addActionListener(new ActionListener() {
	public void actionPerformed(ActionEvent e) {
	  m_History.removeResult(selectedName);
	}
      });
    } else {
      deleteOutput.setEnabled(false);
    }
    resultListMenu.add(deleteOutput);

    resultListMenu.addSeparator();

    JMenuItem loadModel = new JMenuItem("Load model");
    loadModel.addActionListener(new ActionListener() {
	public void actionPerformed(ActionEvent e) {
	  loadClusterer();
	}
      });
    resultListMenu.add(loadModel);

    FastVector o = null;
    if (selectedName != null) {
      o = (FastVector)m_History.getNamedObject(selectedName);
    }

    VisualizePanel temp_vp = null;
    String temp_grph = null;
    Clusterer temp_clusterer = null;
    Instances temp_trainHeader = null;
    int[] temp_ignoreAtts = null;
    
    if (o != null) {
      for (int i = 0; i < o.size(); i++) {
	Object temp = o.elementAt(i);
	if (temp instanceof Clusterer) {
	  temp_clusterer = (Clusterer)temp;
	} else if (temp instanceof Instances) { // training header
	  temp_trainHeader = (Instances)temp;
	} else if (temp instanceof int[]) { // ignored attributes
	  temp_ignoreAtts = (int[])temp;
	} else 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;
    final Clusterer clusterer = temp_clusterer;
    final Instances trainHeader = temp_trainHeader;
    final int[] ignoreAtts = temp_ignoreAtts;
    
    JMenuItem saveModel = new JMenuItem("Save model");
    if (clusterer != null) {
      saveModel.addActionListener(new ActionListener() {
	  public void actionPerformed(ActionEvent e) {
	    saveClusterer(selectedName, clusterer, trainHeader, ignoreAtts);
	  }
	});
    } else {
      saveModel.setEnabled(false);
    }
    resultListMenu.add(saveModel);
    
    JMenuItem reEvaluate =
      new JMenuItem("Re-evaluate model on current test set");
    if (clusterer != null && m_TestInstances != null) {
      reEvaluate.addActionListener(new ActionListener() {
	  public void actionPerformed(ActionEvent e) {
	    reevaluateModel(selectedName, clusterer, trainHeader, ignoreAtts);
	  }
	}); 
    } else {
      reEvaluate.setEnabled(false);
    }
    resultListMenu.add(reEvaluate);
    
    resultListMenu.addSeparator();