attributevisualizationpanel.java

Java 编写的多种数据挖掘算法 包括聚类、分类、预处理等

   * @param index The index of the attribute
   */
  public void setAttribute(int index) {
    
    synchronized (m_locker) {
      m_threadRun = true;
      //if(m_hc!=null && m_hc.isAlive()) m_hc.stop();
      m_attribIndex = index;
      m_as = m_data.attributeStats(m_attribIndex);
      //m_classIndex = m_colorAttrib.getSelectedIndex();
    }
    calcGraph();
  }
  
  /**
   * Recalculates the barplot or histogram to display, required usually when the
   * attribute is changed or the component is resized.
   */
  public void calcGraph() {
    
    synchronized (m_locker) {
      m_threadRun = true;
      if(m_as.nominalCounts!=null) {
        m_hc = new BarCalc();
        m_hc.setPriority(m_hc.MIN_PRIORITY);
        m_hc.start();
      }
      else if(m_as.numericStats!=null) {
        m_hc = new HistCalc();
        m_hc.setPriority(m_hc.MIN_PRIORITY);
        m_hc.start();
      } else {
        m_histBarCounts = null;
        m_histBarClassCounts = null;
        this.repaint();
        m_threadRun = false;
      }
    }
  }
  
  /**
   * Internal class that calculates the barplot to display, in a separate
   * thread. In particular it initializes some of the crucial internal fields
   * required by paintComponent() to display the histogram for the current
   * attribute. These include: m_histBarCounts or m_histBarClassCounts,
   * m_maxValue and m_colorList.
   */
  private class BarCalc extends Thread {
    public void run() {
      synchronized (m_locker) {
        if((m_classIndex >= 0) &&
        (m_data.attribute(m_classIndex).isNominal())) {
          int histClassCounts[][];
          histClassCounts=new int[m_data.attribute(m_attribIndex).numValues()]
                                 [m_data.attribute(m_classIndex).numValues()+1];
          
          m_maxValue = m_as.nominalCounts[0];
          for(int i=0; i<m_data.attribute(m_attribIndex).numValues(); i++) {
            if(m_as.nominalCounts[i]>m_maxValue)
              m_maxValue = m_as.nominalCounts[i];
          }
          
          if(m_colorList.size()==0)
            m_colorList.addElement(Color.black);
          for(int i=m_colorList.size();
          i < m_data.attribute(m_classIndex).numValues()+1; i++) {
            Color pc = m_defaultColors[(i-1) % 10];
            int ija =  (i-1) / 10;
            ija *= 2;
            
            for (int j=0;j<ija;j++) {
              pc = pc.darker();
            }
            
            m_colorList.addElement(pc);
          }
          
          for(int k=0; k<m_data.numInstances(); k++) {
            //System.out.println("attrib: "+
            //                   m_data.instance(k).value(m_attribIndex)+
            //                   " class: "+
            //                   m_data.instance(k).value(m_classIndex));
            if(!m_data.instance(k).isMissing(m_attribIndex))
              if(m_data.instance(k).isMissing(m_classIndex))
                histClassCounts[(int)m_data.instance(k).value(m_attribIndex)]
                               [0]++;
              else
                histClassCounts[(int)m_data.instance(k).value(m_attribIndex)]
                              [(int)m_data.instance(k).value(m_classIndex)+1]++;
          }
          
          //for(int i=0; i<histClassCounts.length; i++) {
          //int sum=0;
          //for(int j=0; j<histClassCounts[i].length; j++) {
          //    sum = sum+histClassCounts[i][j];
          //}
          //System.out.println("histCount: "+sum+" Actual: "+
          //                   m_as.nominalCounts[i]);
          //}
          
          m_threadRun=false;
          m_histBarClassCounts = histClassCounts;
          //Image tmpImg = new BufferedImage(getWidth(), getHeight(),
          //                                 BufferedImage.TYPE_INT_RGB);
          //drawGraph( tmpImg.getGraphics() );
          //img = tmpImg;
          AttributeVisualizationPanel.this.repaint();
        }
        else {
          int histCounts[];
          histCounts  = new int[m_data.attribute(m_attribIndex).numValues()];
          
          m_maxValue = m_as.nominalCounts[0];
          for(int i=0; i<m_data.attribute(m_attribIndex).numValues(); i++) {
            if(m_as.nominalCounts[i]>m_maxValue)
              m_maxValue = m_as.nominalCounts[i];
          }
          
          for(int k=0; k<m_data.numInstances(); k++) {
            if(!m_data.instance(k).isMissing(m_attribIndex))
              histCounts[(int)m_data.instance(k).value(m_attribIndex)]++;
          }
          m_threadRun=false;
          m_histBarCounts = histCounts;
          //Image tmpImg = new BufferedImage(getWidth(), getHeight(),
          //                                 BufferedImage.TYPE_INT_RGB);
          //drawGraph( tmpImg.getGraphics() );
          //img = tmpImg;
          AttributeVisualizationPanel.this.repaint();
        }
      } //end synchronized
    }  //end run()
  }
  
  /**
   * Internal class that calculates the histogram to display, in a separate
   * thread. In particular it initializes some of the crucial internal fields
   * required by paintComponent() to display the histogram for the current
   * attribute. These include: m_histBarCounts or m_histBarClassCounts,
   * m_maxValue and m_colorList.
   */
  private class HistCalc extends Thread {
    public void run() {
      synchronized (m_locker) {
        if((m_classIndex >= 0) &&
           (m_data.attribute(m_classIndex).isNominal())) {
          
          int intervals; double intervalWidth=0.0;
          
          //This uses the M.P.Wand's method to calculate the histogram's
          //interval width. See "Data-Based Choice of Histogram Bin Width", in
          //The American Statistician, Vol. 51, No. 1, Feb., 1997, pp. 59-64.
          //intervalWidth = Math.pow(6D/( -psi(2, g21())*m_data.numInstances()),
          //                          1/3D );
          
          //This uses the Scott's method to calculate the histogram's interval
          //width. See "On optimal and data-based histograms".
          // See Biometrika, 66, 605-610 OR see the same paper mentioned above.
          intervalWidth =  3.49 * m_as.numericStats.stdDev *
                           Math.pow(m_data.numInstances(), -1/3D);
          //The Math.max is introduced to remove the possibility of
          //intervals=0 and =NAN that can happen if respectively all the numeric
          //values are the same or the interval width is evaluated to zero.
          intervals = Math.max(1,
          (int)Math.round( (m_as.numericStats.max - m_as.numericStats.min) /
                           intervalWidth) );
          
          //System.out.println("Max: "+m_as.numericStats.max+
          //                   " Min: "+m_as.numericStats.min+
          //                   " stdDev: "+m_as.numericStats.stdDev+
          //                   "intervalWidth: "+intervalWidth);
          
          //The number 4 below actually represents a padding of 3 pixels on
          //each side of the histogram, and is also reflected in other parts of 
          //the code in the shape of numerical constants like "6" here.
          if(intervals > AttributeVisualizationPanel.this.getWidth()) {
            intervals = AttributeVisualizationPanel.this.getWidth()-6;
            if(intervals<1)//if width is too small then use 1 and forget padding
              intervals = 1;
          }
          int histClassCounts[][]  =
                          new int[intervals]
                                 [m_data.attribute(m_classIndex).numValues()+1];
          
          double barRange   = (m_as.numericStats.max - m_as.numericStats.min) /
                              (double)histClassCounts.length;
          
          m_maxValue = 0;
          
          if(m_colorList.size()==0)
            m_colorList.addElement(Color.black);
          for(int i = m_colorList.size();
          i < m_data.attribute(m_classIndex).numValues()+1; i++) {
            Color pc = m_defaultColors[(i-1) % 10];
            int ija =  (i-1) / 10;
            ija *= 2;
            for (int j=0;j<ija;j++) {
              pc = pc.darker();
            }
            m_colorList.addElement(pc);
          }
          
          for(int k=0; k<m_data.numInstances(); k++) {
            int t=0; //This holds the interval that the attibute value of the
                     //new instance belongs to.
            try {
              if(!m_data.instance(k).isMissing(m_attribIndex)) {
                //1. see footnote at the end of this file
                t = (int)Math.ceil( (float)(
                (m_data.instance(k).value(m_attribIndex)-m_as.numericStats.min)
                / barRange) );
                if(t==0) {
                  if(m_data.instance(k).isMissing(m_classIndex))
                    histClassCounts[t][0]++;
                  else
                    histClassCounts[t][(int)m_data.instance(k).value(m_classIndex)+1]++;
                  //if(histCounts[t]>m_maxValue)
                  //  m_maxValue = histCounts[t];
                }
                else {
                  if(m_data.instance(k).isMissing(m_classIndex))
                    histClassCounts[t-1][0]++;
                  else
                    histClassCounts[t-1][(int)m_data.instance(k).value(m_classIndex)+1]++;
                  //if(histCounts[t-1]>m_maxValue)
                  //  m_maxValue = histCounts[t-1];
                }
              }
            }
            catch(ArrayIndexOutOfBoundsException ae) {
              System.out.println("t:"+(t)+
              " barRange:"+barRange+
              " histLength:"+histClassCounts.length+
              " value:"+m_data.instance(k).value(m_attribIndex)+
              " min:"+m_as.numericStats.min+
              " sumResult:"+
              (m_data.instance(k).value(m_attribIndex) -
              m_as.numericStats.min)+
              " divideResult:"+
              (float)((m_data.instance(k).value(m_attribIndex) -
              m_as.numericStats.min) / barRange)+
              " finalResult:"+
              Math.ceil((float)((m_data.instance(k).value(m_attribIndex)-
              m_as.numericStats.min) / barRange)) );
            }
          }
          for(int i=0; i<histClassCounts.length; i++) {
            int sum=0;
            for(int j=0; j<histClassCounts[i].length; j++)
              sum = sum+histClassCounts[i][j];
            if(m_maxValue<sum)
              m_maxValue = sum;
          }
          m_histBarClassCounts = histClassCounts;
          m_barRange =  barRange;
          
        }
        else { //else if the class attribute is numeric or the class is not set
          
          int intervals; double intervalWidth;
          //At the time of this coding the
          //possibility of datasets with zero instances
          //was being dealt with in the
          //PreProcessPanel of weka Explorer.
          
          //old method of calculating number of intervals
          //intervals =  m_as.totalCount>10 ?
          //                  (int)(m_as.totalCount*0.1):(int)m_as.totalCount;
          
          //This uses the M.P.Wand's method to calculate the histogram's
          //interval width. See "Data-Based Choice of Histogram Bin Width", in
          //The American Statistician, Vol. 51, No. 1, Feb., 1997, pp. 59-64.
          //intervalWidth = Math.pow(6D/(-psi(2, g21())*m_data.numInstances() ),
          //                          1/3D );
          
          //This uses the Scott's method to calculate the histogram's interval
          //width. See "On optimal and data-based histograms".
          // See Biometrika, 66, 605-610 OR see the same paper mentioned above.
          intervalWidth =  3.49 * m_as.numericStats.stdDev *
                           Math.pow(m_data.numInstances(), -1/3D);
          //The Math.max is introduced to remove the possibility of
          //intervals=0 and =NAN that can happen if respectively all the numeric
          //values are the same or the interval width is evaluated to zero.
          intervals = Math.max(1,
          (int)Math.round( (m_as.numericStats.max - m_as.numericStats.min) /
                           intervalWidth) );
          
          //The number 4 below actually represents a padding of 3 pixels on
          //each side of the histogram, and is also reflected in other parts of 
          //the code in the shape of numerical constants like "6" here.
          if(intervals > AttributeVisualizationPanel.this.getWidth()) {
            intervals = AttributeVisualizationPanel.this.getWidth()-6;
            if(intervals<1)
              intervals = 1;
          }
          
          int histCounts[]  = new int[intervals];
          double barRange   = (m_as.numericStats.max - m_as.numericStats.min) /
                              (double)histCounts.length;
          
          m_maxValue = 0;
          
          for(int k=0; k<m_data.numInstances(); k++) {
            int t=0; //This holds the interval to which the current attribute's
                    //value of this particular instance k belongs to.
            
            if(m_data.instance(k).isMissing(m_attribIndex)) 
              continue; //ignore missing values
            
            try {
              //1. see footnote at the end of this file
              t =(int) Math.ceil((float)(
              (m_data.instance(k).value(m_attribIndex)-m_as.numericStats.min)
              / barRange));
              if(t==0) {
                histCounts[t]++;
                if(histCounts[t]>m_maxValue)
                  m_maxValue = histCounts[t];
              }
              else {
                histCounts[t-1]++;
                if(histCounts[t-1]>m_maxValue)
                  m_maxValue = histCounts[t-1];
              }
            }
            catch(ArrayIndexOutOfBoundsException ae) {
              ae.printStackTrace();
              System.out.println("t:"+(t)+
              " barRange:"+barRange+
              " histLength:"+histCounts.length+
              " value:"+m_data.instance(k).value(m_attribIndex)+