boundarypanel.java

一个数据挖掘软件ALPHAMINERR的整个过程的JAVA版源代码

   * @exception Exception if an error occurs
   */
  public void start() throws Exception {
    m_numOfSamplesPerGenerator = 
      (int)Math.pow(m_samplesBase, m_trainingData.numAttributes()-3);

    m_stopReplotting = true;
    if (m_trainingData == null) {
      throw new Exception("No training data set (BoundaryPanel)");
    }
    if (m_classifier == null) {
      throw new Exception("No classifier set (BoundaryPanel)");
    }
    if (m_dataGenerator == null) {
      throw new Exception("No data generator set (BoundaryPanel)");
    }
    if (m_trainingData.attribute(m_xAttribute).isNominal() || 
	m_trainingData.attribute(m_yAttribute).isNominal()) {
      throw new Exception("Visualization dimensions must be numeric "
			  +"(BoundaryPanel)");
    }
    
    computeMinMaxAtts();
    
    if (m_plotThread == null) {
      m_plotThread = new PlotThread();
      m_plotThread.setPriority(Thread.MIN_PRIORITY);
      m_plotThread.start();
    }
  }
  
  // Thread for main plotting operation
  protected class PlotThread extends Thread {
    double [] m_weightingAttsValues;
    boolean [] m_attsToWeightOn;
    double [] m_vals;
    double [] m_dist;
    Instance m_predInst;
    public void run() {

      m_stopPlotting = false;
      try {
        initialize();
        repaint();
        
        // train the classifier
        m_probabilityCache = new double[m_panelHeight][m_panelWidth][];
        m_classifier.buildClassifier(m_trainingData);
        
        // build DataGenerator
        m_attsToWeightOn = new boolean[m_trainingData.numAttributes()];
        m_attsToWeightOn[m_xAttribute] = true;
        m_attsToWeightOn[m_yAttribute] = true;
	      
        m_dataGenerator.setWeightingDimensions(m_attsToWeightOn);
	      
        m_dataGenerator.buildGenerator(m_trainingData);

        // generate samples
        m_weightingAttsValues = new double [m_attsToWeightOn.length];
        m_vals = new double[m_trainingData.numAttributes()];
        m_predInst = new Instance(1.0, m_vals);
        m_predInst.setDataset(m_trainingData);

	
        m_size = 1 << 4;  // Current sample region size
	
	m_initialTiling = true;
        // Display the initial coarse image tiling.
      abortInitial:
        for (int i = 0; i <= m_panelHeight; i += m_size) {   
          for (int j = 0; j <= m_panelWidth; j += m_size) {   
	    if (m_stopPlotting) {
	      break abortInitial;
	    }
	    if (m_pausePlotting) {
	      synchronized (m_dummy) {
		try {
		  m_dummy.wait();
		} catch (InterruptedException ex) {
		  m_pausePlotting = false;
		}
	      }
	    }
            plotPoint(j, i, m_size, m_size, 
		      calculateRegionProbs(j, i), (j == 0));
          }
        }
	if (!m_stopPlotting) {
	  m_initialTiling = false;
	}
        
        // Sampling and gridding loop
        int size2 = m_size / 2;
        abortPlot: 
        while (m_size > 1) { // Subdivide down to the pixel level
          for (int i = 0; i <= m_panelHeight; i += m_size) {
            for (int j = 0; j <= m_panelWidth; j += m_size) {
              if (m_stopPlotting) {
                break abortPlot;
              }
	      if (m_pausePlotting) {
		synchronized (m_dummy) {
		  try {
		    m_dummy.wait();
		  } catch (InterruptedException ex) {
		    m_pausePlotting = false;
		  }
		}
	      }
              boolean update = (j == 0 && i % 2 == 0);
              // Draw the three new subpixel regions
              plotPoint(j, i + size2, size2, size2, 
			calculateRegionProbs(j, i + size2), update);
              plotPoint(j + size2, i + size2, size2, size2, 
			calculateRegionProbs(j + size2, i + size2), update);
              plotPoint(j + size2, i, size2, size2, 
			calculateRegionProbs(j + size2, i), update);
            }
          }
          // The new region edge length is half the old edge length
          m_size = size2;
          size2 = size2 / 2;
        }
	update();
	

	/*
        // Old method without sampling.
        abortPlot: 
        for (int i = 0; i < m_panelHeight; i++) {
          for (int j = 0; j < m_panelWidth; j++) {
            if (m_stopPlotting) {
              break abortPlot;
            }
            plotPoint(j, i, calculateRegionProbs(j, i), (j == 0));
          }
        }
        */


        if (m_plotTrainingData) {
          plotTrainingData();
        }
	      
      } catch (Exception ex) {
        ex.printStackTrace();
      } finally {
        m_plotThread = null;
        // notify any listeners that we are finished
        Vector l;
        ActionEvent e = new ActionEvent(this, 0, "");
        synchronized(this) {
          l = (Vector)m_listeners.clone();
        }
        for (int i = 0; i < l.size(); i++) {
          ActionListener al = (ActionListener)l.elementAt(i);
          al.actionPerformed(e);
        }
      }
    }
    
    private double [] calculateRegionProbs(int j, int i) throws Exception {
      double [] sumOfProbsForRegion = 
	new double [m_trainingData.classAttribute().numValues()];

      for (int u = 0; u < m_numOfSamplesPerRegion; u++) {
      
        double [] sumOfProbsForLocation = 
	  new double [m_trainingData.classAttribute().numValues()];
      
        m_weightingAttsValues[m_xAttribute] = getRandomX(j);
        m_weightingAttsValues[m_yAttribute] = getRandomY(m_panelHeight-i-1);
      
        m_dataGenerator.setWeightingValues(m_weightingAttsValues);
      
        double [] weights = m_dataGenerator.getWeights();
        double sumOfWeights = Utils.sum(weights);
        int [] indices = Utils.sort(weights);
      
        // Prune 1% of weight mass
        int [] newIndices = new int[indices.length];
        double sumSoFar = 0; 
        double criticalMass = 0.99 * sumOfWeights;
        int index = weights.length - 1; int counter = 0;
        for (int z = weights.length - 1; z >= 0; z--) {
          newIndices[index--] = indices[z];
          sumSoFar += weights[indices[z]];
          counter++;
          if (sumSoFar > criticalMass) {
            break;
          }
        }
        indices = new int[counter];
        System.arraycopy(newIndices, index + 1, indices, 0, counter);
      
        for (int z = 0; z < m_numOfSamplesPerGenerator; z++) {
        
          m_dataGenerator.setWeightingValues(m_weightingAttsValues);
          double [][] values = m_dataGenerator.generateInstances(indices);
        
          for (int q = 0; q < values.length; q++) {
            if (values[q] != null) {
              System.arraycopy(values[q], 0, m_vals, 0, m_vals.length);
              m_vals[m_xAttribute] = m_weightingAttsValues[m_xAttribute];
              m_vals[m_yAttribute] = m_weightingAttsValues[m_yAttribute];
            
              // classify the instance
              m_dist = m_classifier.distributionForInstance(m_predInst);
              for (int k = 0; k < sumOfProbsForLocation.length; k++) {
                sumOfProbsForLocation[k] += (m_dist[k] * weights[q]); 
              }
            }
          }
        }
      
        for (int k = 0; k < sumOfProbsForRegion.length; k++) {
          sumOfProbsForRegion[k] += (sumOfProbsForLocation[k] * 
				     sumOfWeights); 
        }
      }
    
      // average
      Utils.normalize(sumOfProbsForRegion);

      // cache
      if ((i < m_panelHeight) && (j < m_panelWidth)) {
        m_probabilityCache[i][j] = new double[sumOfProbsForRegion.length];
        System.arraycopy(sumOfProbsForRegion, 0, m_probabilityCache[i][j], 
			 0, sumOfProbsForRegion.length);
      }
		
      return sumOfProbsForRegion;
    }
  }

  protected void plotTrainingData() {
    Graphics2D osg = (Graphics2D)m_osi.getGraphics();
    Graphics g = m_plotPanel.getGraphics();
    osg.setRenderingHint(RenderingHints.KEY_ANTIALIASING,
                         RenderingHints.VALUE_ANTIALIAS_ON);
    double xval = 0; double yval = 0;
    for (int i = 0; i < m_trainingData.numInstances(); i++) {
      if (!m_trainingData.instance(i).isMissing(m_xAttribute) &&
          !m_trainingData.instance(i).isMissing(m_yAttribute)) {

        xval = m_trainingData.instance(i).value(m_xAttribute);
        yval = m_trainingData.instance(i).value(m_yAttribute);
       
        int panelX = convertToPanelX(xval);
        int panelY = convertToPanelY(yval);
        Color ColorToPlotWith = 
          ((Color)m_Colors.elementAt((int)m_trainingData.instance(i).
                                     value(m_classIndex) % m_Colors.size()));

        if (ColorToPlotWith.equals(Color.white)) {
          osg.setColor(Color.black);
        } else {
          osg.setColor(Color.white);
        }
        osg.fillOval(panelX-3, panelY-3, 7, 7);
        osg.setColor(ColorToPlotWith);
        osg.fillOval(panelX-2, panelY-2, 5, 5);
      }
    }
    g.drawImage(m_osi,0,0,m_plotPanel);
  }

  private int convertToPanelX(double xval) {
    double temp = (xval - m_minX) / m_rangeX;
    temp = temp * (double) m_panelWidth;

    return (int)temp;
  }

  private int convertToPanelY(double yval) {
    double temp = (yval - m_minY) / m_rangeY;
    temp = temp * (double) m_panelHeight;
    temp = m_panelHeight - temp;
    
    return (int)temp;
  }
  
  private double convertFromPanelX(double pX) {
    pX /= (double) m_panelWidth;
    
    pX *= m_rangeX;
    return pX + m_minX;
  }

  private double convertFromPanelY(double pY) {
    pY  = m_panelHeight - pY;
    pY /= (double) m_panelHeight;
    pY *= m_rangeY;
    
    return pY + m_minY;
  }


  protected  void plotPoint(int x, int y, double [] probs, boolean update) {
    plotPoint(x, y, 1, 1, probs, update);
  }
  
  private void plotPoint(int x, int y, int width, int height, 
			 double [] probs, boolean update) {

    // draw a progress line
    Graphics osg = m_osi.getGraphics();
    if (update) {
      osg.setXORMode(Color.white);
      osg.drawLine(0, y, m_panelWidth-1, y);
      update();
      osg.drawLine(0, y, m_panelWidth-1, y);
    }

    // plot the point
    osg.setPaintMode();
    float [] colVal = new float[3];
    
    float [] tempCols = new float[3];
    for (int k = 0; k < probs.length; k++) {
      Color curr = (Color)m_Colors.elementAt(k % m_Colors.size());

      curr.getRGBColorComponents(tempCols);
      for (int z = 0 ; z < 3; z++) {
        colVal[z] += probs[k] * tempCols[z];
      }
    }

    for (int z = 0; z < 3; z++) {
      if (colVal[z] < 0) {
	colVal[z] = 0;
      } else if (colVal[z] > 1) {
	colVal[z] = 1;
      }
    }
    
    osg.setColor(new Color(colVal[0], 
                           colVal[1], 
                           colVal[2]));
    osg.fillRect(x, y, width, height);