simplekmeans.java

数据挖掘聚类算法：SimpleKMeans源代码

  public String toString() {
    if (m_ClusterCentroids == null) {
      return "No clusterer built yet!";
    }

    int maxWidth = 0;
    int maxAttWidth = 0;
    boolean containsNumeric = false;
    for (int i = 0; i < m_NumClusters; i++) {
      for (int j = 0 ;j < m_ClusterCentroids.numAttributes(); j++) {
        if (m_ClusterCentroids.attribute(j).name().length() > maxAttWidth) {
          maxAttWidth = m_ClusterCentroids.attribute(j).name().length();
        }
	if (m_ClusterCentroids.attribute(j).isNumeric()) {
          containsNumeric = true;
	  double width = Math.log(Math.abs(m_ClusterCentroids.instance(i).value(j))) /
	    Math.log(10.0);
          //          System.err.println(m_ClusterCentroids.instance(i).value(j)+" "+width);
          if (width < 0) {
            width = 1;
          }
          // decimal + # decimal places + 1
	  width += 6.0;
	  if ((int)width > maxWidth) {
	    maxWidth = (int)width;
	  }
	}
      }
    }

    for (int i = 0; i < m_ClusterCentroids.numAttributes(); i++) {
      if (m_ClusterCentroids.attribute(i).isNominal()) {
        Attribute a = m_ClusterCentroids.attribute(i);
        for (int j = 0; j < m_ClusterCentroids.numInstances(); j++) {
          String val = a.value((int)m_ClusterCentroids.instance(j).value(i));
          if (val.length() > maxWidth) {
            maxWidth = val.length();
          }
        }
        for (int j = 0; j < a.numValues(); j++) {
          String val = a.value(j) + " ";
          if (val.length() > maxAttWidth) {
            maxAttWidth = val.length();
          }
        }
      }
    }

    if (m_displayStdDevs) {
      // check for maximum width of maximum frequency count
      for (int i = 0; i < m_ClusterCentroids.numAttributes(); i++) {
        if (m_ClusterCentroids.attribute(i).isNominal()) {
          int maxV = Utils.maxIndex(m_FullNominalCounts[i]);
          /*          int percent = (int)((double)m_FullNominalCounts[i][maxV] /
                      Utils.sum(m_ClusterSizes) * 100.0); */
          int percent = 6; // max percent width (100%)
          String nomV = "" + m_FullNominalCounts[i][maxV];
            //            + " (" + percent + "%)";
          if (nomV.length() + percent > maxWidth) {
            maxWidth = nomV.length() + 1;
          }
        }
      }
    }

    // check for size of cluster sizes
    for (int i = 0; i < m_ClusterSizes.length; i++) {
      String size = "(" + m_ClusterSizes[i] + ")";
      if (size.length() > maxWidth) {
        maxWidth = size.length();
      }
    }
    
    if (m_displayStdDevs && maxAttWidth < "missing".length()) {
      maxAttWidth = "missing".length();
    }
    
    String plusMinus = "+/-";
    maxAttWidth += 2;
    if (m_displayStdDevs && containsNumeric) {
      maxWidth += plusMinus.length();
    }
    if (maxAttWidth < "Attribute".length() + 2) {
      maxAttWidth = "Attribute".length() + 2;
    }

    if (maxWidth < "Full Data".length()) {
      maxWidth = "Full Data".length() + 1;
    }

    if (maxWidth < "missing".length()) {
      maxWidth = "missing".length() + 1;
    }


    
    StringBuffer temp = new StringBuffer();
    //    String naString = "N/A";

    
    /*    for (int i = 0; i < maxWidth+2; i++) {
      naString += " ";
      } */
    temp.append("\nkMeans\n======\n");
    temp.append("\nNumber of iterations: " + m_Iterations+"\n");
    temp.append("Within cluster sum of squared errors: " + Utils.sum(m_squaredErrors));
    if (!m_dontReplaceMissing) {
      temp.append("\nMissing values globally replaced with mean/mode");
    }

    temp.append("\n\nCluster centroids:\n");
    temp.append(pad("Cluster#", " ", (maxAttWidth + (maxWidth * 2 + 2)) - "Cluster#".length(), true));

    temp.append("\n");
    temp.append(pad("Attribute", " ", maxAttWidth - "Attribute".length(), false));

    
    temp.append(pad("Full Data", " ", maxWidth + 1 - "Full Data".length(), true));

    // cluster numbers
    for (int i = 0; i < m_NumClusters; i++) {
      String clustNum = "" + i;
      temp.append(pad(clustNum, " ", maxWidth + 1 - clustNum.length(), true));
    }
    temp.append("\n");

    // cluster sizes
    String cSize = "(" + Utils.sum(m_ClusterSizes) + ")";
    temp.append(pad(cSize, " ", maxAttWidth + maxWidth + 1 - cSize.length(), true));
    for (int i = 0; i < m_NumClusters; i++) {
      cSize = "(" + m_ClusterSizes[i] + ")";
      temp.append(pad(cSize, " ",maxWidth + 1 - cSize.length(), true));
    }
    temp.append("\n");

    temp.append(pad("", "=", maxAttWidth + 
                    (maxWidth * (m_ClusterCentroids.numInstances()+1) 
                     + m_ClusterCentroids.numInstances() + 1), true));
    temp.append("\n");

    for (int i = 0; i < m_ClusterCentroids.numAttributes(); i++) {
      String attName = m_ClusterCentroids.attribute(i).name();
      temp.append(attName);
      for (int j = 0; j < maxAttWidth - attName.length(); j++) {
        temp.append(" ");
      }

      String strVal;
      String valMeanMode;
      // full data
      if (m_ClusterCentroids.attribute(i).isNominal()) {
        if (m_FullMeansOrModes[i] == -1) { // missing
          valMeanMode = pad("missing", " ", maxWidth + 1 - "missing".length(), true);
        } else {
          valMeanMode = 
            pad((strVal = m_ClusterCentroids.attribute(i).value((int)m_FullMeansOrModes[i])),
                " ", maxWidth + 1 - strVal.length(), true);
        }
      } else {
        if (Double.isNaN(m_FullMeansOrModes[i])) {
          valMeanMode = pad("missing", " ", maxWidth + 1 - "missing".length(), true);
        } else {
          valMeanMode =  pad((strVal = Utils.doubleToString(m_FullMeansOrModes[i],
                                                            maxWidth,4).trim()), 
                             " ", maxWidth + 1 - strVal.length(), true);
        }
      }
      temp.append(valMeanMode);

      for (int j = 0; j < m_NumClusters; j++) {
        if (m_ClusterCentroids.attribute(i).isNominal()) {
          if (m_ClusterCentroids.instance(j).isMissing(i)) {
            valMeanMode = pad("missing", " ", maxWidth + 1 - "missing".length(), true);
          } else {
            valMeanMode = 
              pad((strVal = m_ClusterCentroids.attribute(i).value((int)m_ClusterCentroids.instance(j).value(i))),
                  " ", maxWidth + 1 - strVal.length(), true);
          }
        } else {
          if (m_ClusterCentroids.instance(j).isMissing(i)) {
            valMeanMode = pad("missing", " ", maxWidth + 1 - "missing".length(), true);
          } else {
            valMeanMode = pad((strVal = Utils.doubleToString(m_ClusterCentroids.instance(j).value(i),
                                                               maxWidth,4).trim()), 
                                " ", maxWidth + 1 - strVal.length(), true);
          }
        }
        temp.append(valMeanMode);
      }
      temp.append("\n");

      if (m_displayStdDevs) {
        // Std devs/max nominal
        String stdDevVal = "";

        if (m_ClusterCentroids.attribute(i).isNominal()) {
          // Do the values of the nominal attribute
          Attribute a = m_ClusterCentroids.attribute(i);
          for (int j = 0; j < a.numValues(); j++) {
            // full data
            String val = "  " + a.value(j);
            temp.append(pad(val, " ", maxAttWidth + 1 - val.length(), false));
            int count = m_FullNominalCounts[i][j];
            int percent = (int)((double)m_FullNominalCounts[i][j] /
                              Utils.sum(m_ClusterSizes) * 100.0);
            String percentS = "" + percent + "%)";
            percentS = pad(percentS, " ", 5 - percentS.length(), true);
            stdDevVal = "" + count + " (" + percentS;
            stdDevVal = 
              pad(stdDevVal, " ", maxWidth + 1 - stdDevVal.length(), true);
            temp.append(stdDevVal);

            // Clusters
            for (int k = 0; k < m_NumClusters; k++) {
              count = m_ClusterNominalCounts[k][i][j];
              percent = (int)((double)m_ClusterNominalCounts[k][i][j] /
                              m_ClusterSizes[k] * 100.0);
              percentS = "" + percent + "%)";
              percentS = pad(percentS, " ", 5 - percentS.length(), true);
              stdDevVal = "" + count + " (" + percentS;
              stdDevVal = 
                pad(stdDevVal, " ", maxWidth + 1 - stdDevVal.length(), true);
              temp.append(stdDevVal);
            }
            temp.append("\n");
          }
          // missing (if any)
          if (m_FullMissingCounts[i] > 0) {
            // Full data
            temp.append(pad("  missing", " ", maxAttWidth + 1 - "  missing".length(), false));
            int count = m_FullMissingCounts[i];
            int percent = (int)((double)m_FullMissingCounts[i] /
                              Utils.sum(m_ClusterSizes) * 100.0);
            String percentS = "" + percent + "%)";
            percentS = pad(percentS, " ", 5 - percentS.length(), true);
            stdDevVal = "" + count + " (" + percentS;
            stdDevVal = 
              pad(stdDevVal, " ", maxWidth + 1 - stdDevVal.length(), true);
            temp.append(stdDevVal);
           
            // Clusters
            for (int k = 0; k < m_NumClusters; k++) {
              count = m_ClusterMissingCounts[k][i];
              percent = (int)((double)m_ClusterMissingCounts[k][i] /
                              m_ClusterSizes[k] * 100.0);
              percentS = "" + percent + "%)";
              percentS = pad(percentS, " ", 5 - percentS.length(), true);
              stdDevVal = "" + count + " (" + percentS;
              stdDevVal = 
                pad(stdDevVal, " ", maxWidth + 1 - stdDevVal.length(), true);
              temp.append(stdDevVal);
            }

            temp.append("\n");
          }

          temp.append("\n");
        } else {
          // Full data
          if (Double.isNaN(m_FullMeansOrModes[i])) {
            stdDevVal = pad("--", " ", maxAttWidth + maxWidth + 1 - 2, true);
          } else {
            stdDevVal = pad((strVal = plusMinus 
                             + Utils.doubleToString(m_FullStdDevs[i],
                                                    maxWidth,4).trim()), 
                            " ", maxWidth + maxAttWidth + 1 - strVal.length(), true);
          }
          temp.append(stdDevVal);

          // Clusters
          for (int j = 0; j < m_NumClusters; j++) {
            if (m_ClusterCentroids.instance(j).isMissing(i)) {
              stdDevVal = pad("--", " ", maxWidth + 1 - 2, true);
            } else {
              stdDevVal = 
                pad((strVal = plusMinus 
                     + Utils.doubleToString(m_ClusterStdDevs.instance(j).value(i),
                                            maxWidth,4).trim()), 
                    " ", maxWidth + 1 - strVal.length(), true);
            }
            temp.append(stdDevVal);
          }
          temp.append("\n\n");
        }
      }
    }

    temp.append("\n\n");
    return temp.toString();
  }

  private String pad(String source, String padChar, 
                     int length, boolean leftPad) {
    StringBuffer temp = new StringBuffer();

    if (leftPad) {
      for (int i = 0; i< length; i++) {
        temp.append(padChar);
      }
      temp.append(source);
    } else {
      temp.append(source);
      for (int i = 0; i< length; i++) {
        temp.append(padChar);
      }
    }
    return temp.toString();
  }

  /**
   * Gets the the cluster centroids
   * 
   * @return		the cluster centroids
   */
  public Instances getClusterCentroids() {
    return m_ClusterCentroids;
  }

  /**
   * Gets the standard deviations of the numeric attributes in each cluster
   * 
   * @return		the standard deviations of the numeric attributes 
   * 			in each cluster
   */
  public Instances getClusterStandardDevs() {
    return m_ClusterStdDevs;
  }

  /**
   * Returns for each cluster the frequency counts for the values of each 
   * nominal attribute
   * 
   * @return		the counts
   */
  public int [][][] getClusterNominalCounts() {
    return m_ClusterNominalCounts;
  }

  /**
   * Gets the squared error for all clusters
   * 
   * @return		the squared error
   */
  public double getSquaredError() {
    return Utils.sum(m_squaredErrors);
  }

  /**
   * Gets the number of instances in each cluster
   * 
   * @return		The number of instances in each cluster
   */
  public int [] getClusterSizes() {
    return m_ClusterSizes;
  }
  
  /**
   * Returns the revision string.
   * 
   * @return		the revision
   */
  public String getRevision() {
    return RevisionUtils.extract("$Revision: 1.39 $");
  }

  /**
   * Main method for testing this class.
   *
   * @param argv should contain the following arguments: <p>
   * -t training file [-N number of clusters]
   */
  public static void main (String[] argv) {
    runClusterer(new SimpleKMeans(), argv);
  }
}