subspacecluster.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

    // check whether all the attributes are covered
    count = new int[getNumAttributes()];
    for (i = 0; i < getNumAttributes(); i++) {
      for (n = 0; n < getClusters().length; n++) {
        cl = (SubspaceClusterDefinition) getClusters()[n];
        r  = new Range(cl.getAttrIndexRange());
        r.setUpper(getNumAttributes());
        if (r.isInRange(i))
          count[i]++;
      }
    }

    // list all indices that are not covered
    attrIndex = "";
    for (i = 0; i < count.length; i++) {
      if (count[i] == 0) {
        if (attrIndex.length() != 0)
          attrIndex += ",";
        attrIndex += (i+1);
      }
    }

    if (attrIndex.length() != 0)
      throw new IllegalArgumentException(
          "The following attributes are not covered by a cluster "
          + "definition: " + attrIndex + "\n");

    return true;
  }

  /**
   * Gets the single mode flag.
   *
   * @return true if methode generateExample can be used.
   */
  public boolean getSingleModeFlag() { 
    return false; 
  }

  /**
   * Initializes the format for the dataset produced. 
   *
   * @return the output data format
   * @throws Exception data format could not be defined 
   */

  public Instances defineDataFormat() throws Exception {

    // initialize
    setOptions(getOptions());

    checkCoverage();

    Random random = new Random (getSeed());
    setRandom(random);
    Instances dataset;
    FastVector attributes = new FastVector(3);
    Attribute attribute;
    boolean classFlag = getClassFlag();

    FastVector classValues = null;
    if (classFlag) 
      classValues = new FastVector(getClusters().length);     
    FastVector boolValues = new FastVector(2);
    boolValues.addElement("false");
    boolValues.addElement("true");
    FastVector nomValues = null;

    // define dataset
    for (int i = 0; i < getNumAttributes(); i++) {
      // define boolean attribute
      if (m_booleanCols.isInRange(i)) {
        attribute = new Attribute("B" + i, boolValues);
      } 
      else if (m_nominalCols.isInRange(i)) {
        // define nominal attribute
        nomValues = new FastVector(m_numValues[i]);
        for (int j = 0; j < m_numValues[i]; j++)
          nomValues.addElement("value-" + j);
        attribute = new Attribute("N" + i, nomValues);
      } 
      else {
        // numerical attribute
        attribute = new Attribute("X" + i); 
      }
      attributes.addElement(attribute);
    }

    if (classFlag) {
      for (int i = 0; i < getClusters().length; i++)
        classValues.addElement("c" + i);
      attribute = new Attribute ("class", classValues); 
      attributes.addElement(attribute);
    }

    dataset = new Instances(getRelationNameToUse(), attributes, 0);
    if (classFlag) 
      dataset.setClassIndex(m_NumAttributes);

    // set dataset format of this class
    Instances format = new Instances(dataset, 0);
    setDatasetFormat(format);

    for (int i = 0; i < getClusters().length; i++) {
      SubspaceClusterDefinition cl = (SubspaceClusterDefinition) getClusters()[i];
      cl.setNumInstances(random);
      cl.setParent(this);
    }

    return dataset; 
  }

  /**
   * Returns true if attribute is boolean
   *@param index of the attribute
   *@return true if the attribute is boolean
   */
  public boolean isBoolean(int index) {
    return m_booleanCols.isInRange(index); 
  }

  /**
   * Returns true if attribute is nominal
   *@param index of the attribute
   *@return true if the attribute is nominal
   */
  public boolean isNominal(int index) {
    return m_nominalCols.isInRange(index);
  }

  /**
   * returns array that stores the number of values for a nominal attribute.
   * 
   * @return the array that stores the number of values for a nominal attribute
   */
  public int[] getNumValues() {
    return m_numValues;
  }

  /**
   * Generate an example of the dataset. 
   * @return the instance generated
   * @throws Exception if format not defined or generating <br/>
   * examples one by one is not possible, because voting is chosen
   */

  public Instance generateExample() throws Exception {
    throw new Exception("Examples cannot be generated one by one.");
  }

  /**
   * Generate all examples of the dataset. 
   * @return the instance generated
   * @throws Exception if format not defined 
   */

  public Instances generateExamples() throws Exception {
    Instances format = getDatasetFormat();
    Instance example = null;

    if (format == null) 
      throw new Exception("Dataset format not defined.");

    // generate examples for one cluster after another
    for (int cNum = 0; cNum < getClusters().length; cNum++) {
      SubspaceClusterDefinition cl  = (SubspaceClusterDefinition) getClusters()[cNum];

      //get the number of instances to create
      int instNum = cl.getNumInstances();

      //class value is c + cluster number
      String cName = "c" + cNum;

      switch (cl.getClusterType().getSelectedTag().getID()) {
        case (UNIFORM_RANDOM):
          for (int i = 0; i < instNum; i++) {
            // generate example
            example = generateExample(format, getRandom(), cl, cName);
            if (example != null)
              format.add(example);
          }
          break;
        case (TOTAL_UNIFORM):
          // generate examples
          if (!cl.isInteger())
            generateUniformExamples(format, instNum, cl, cName);
          else
            generateUniformIntegerExamples(format, instNum, cl, cName);
          break;
        case (GAUSSIAN):
          // generate examples
          generateGaussianExamples(format, instNum, getRandom(), cl, cName);
          break;
      }
    }

    return format;
  }

  /**
   * Generate an example of the dataset. 
   * 
   * @param format the dataset format
   * @param randomG the random number generator to use
   * @param cl the cluster definition
   * @param cName the class value
   * @return the generated instance
   */
  private Instance generateExample(
      Instances format, Random randomG, SubspaceClusterDefinition cl, 
      String cName) {

    boolean makeInteger = cl.isInteger();
    int num = -1;
    Instance example = null;
    int numAtts = m_NumAttributes;
    if (getClassFlag()) numAtts++;

    example = new Instance(numAtts);
    example.setDataset(format);
    boolean[] attributes = cl.getAttributes();
    double[] minValue = cl.getMinValue();
    double[] maxValue = cl.getMaxValue();
    double value;

    int clusterI = -1;
    for (int i = 0; i < m_NumAttributes; i++) {
      if (attributes[i]) {
        clusterI++;
        num++;
        // boolean  or nominal attribute
        if (isBoolean(i) || isNominal(i)) {

          if (minValue[clusterI] == maxValue[clusterI]) {
            value = minValue[clusterI];
          } 
          else {
            int numValues = (int)(maxValue[clusterI] - minValue[clusterI] + 1.0);
            value = randomG.nextInt(numValues);
            value += minValue[clusterI];
          }
        } 
        else {
          // numeric attribute
          value = randomG.nextDouble() * 
            (maxValue[num] - minValue[num]) + minValue[num];
          if (makeInteger)
            value = Math.round(value);
        }
        example.setValue(i, value);
      } 
      else {
        example.setMissing(i);
      }
    }

    if (getClassFlag())
      example.setClassValue(cName);

    return example; 
  }

  /**
   * Generate examples for a uniform cluster dataset. 
   * 
   * @param format the dataset format
   * @param numInstances the number of instances to generator
   * @param cl the cluster definition
   * @param cName the class value
   */
  private void generateUniformExamples(
      Instances format, int numInstances, SubspaceClusterDefinition cl, 
      String cName) {

    Instance example = null;
    int numAtts = m_NumAttributes;
    if (getClassFlag()) numAtts++;

    example = new Instance(numAtts);
    example.setDataset(format);
    boolean[] attributes = cl.getAttributes();
    double[] minValue = cl.getMinValue();
    double[] maxValue = cl.getMaxValue();
    double[] diff = new double[minValue.length];

    for (int i = 0; i < minValue.length; i++)
      diff[i] = (maxValue[i] - minValue[i]);

    for (int j = 0; j < numInstances; j++) {
      int num = -1;
      for (int i = 0; i < m_NumAttributes; i++) {
        if (attributes[i]) {
          num++;
          double value = minValue[num] + (diff[num] * (double)((double)j / (double)(numInstances - 1)));
          example.setValue(i, value);
        } 
        else {
          example.setMissing(i);
        }
      }
      if (getClassFlag())
        example.setClassValue(cName);
      format.add(example);
    }
  }

  /**
   * Generate examples for a uniform cluster dataset. 
   * 
   * @param format the dataset format
   * @param numInstances the number of instances to generator
   * @param cl the cluster definition
   * @param cName the class value
   */
  private void generateUniformIntegerExamples(
      Instances format, int numInstances, SubspaceClusterDefinition cl, 
      String cName) {

    Instance example = null;
    int numAtts = m_NumAttributes;
    if (getClassFlag()) numAtts++;

    example = new Instance(numAtts);
    example.setDataset(format);
    boolean[] attributes = cl.getAttributes();
    double[] minValue = cl.getMinValue();
    double[] maxValue = cl.getMaxValue();
    int[] minInt = new int[minValue.length];
    int[] maxInt = new int[maxValue.length];
    int[] intValue = new int[maxValue.length];
    int[] numInt = new int[minValue.length];

    int num = 1;
    for (int i = 0; i < minValue.length; i++) {
      minInt[i] = (int)Math.ceil(minValue[i]);
      maxInt[i] = (int)Math.floor(maxValue[i]);
      numInt[i] = (maxInt[i] - minInt[i] + 1);
      num = num * numInt[i];
    }
    int numEach = numInstances / num;
    int rest = numInstances - numEach * num;

    // initialize with smallest values combination
    for (int i = 0; i < m_NumAttributes; i++) {
      if (attributes[i]) {
        example.setValue(i, (double)minInt[i]);
        intValue[i] = minInt[i];
      } 
      else {
        example.setMissing(i);
      }
    }
    if (getClassFlag())
      example.setClassValue(cName);
    int added = 0;
    int attr = 0;
    // do while not added all
    do {
      // add all for one value combination
      for (int k = 0; k < numEach; k++) {
        format.add(example);
        example = (Instance) example.copy();
        added++;
      }
      if (rest > 0) {
        format.add(example);
        example = (Instance) example.copy();
        added++;
        rest--;
      }

      if (added >= numInstances) break;
      // switch to the next value combination
      boolean done = false;
      do {
        if (attributes[attr] && (intValue[attr] + 1 <= maxInt[attr])) {
          intValue[attr]++;
          done = true;
        } 
        else {
          attr++;
        }
      } while (!done);

      example.setValue(attr, (double)intValue[attr]);
    } while (added < numInstances);
  }

  /**
   * Generate examples for a uniform cluster dataset. 
   * 
   * @param format the dataset format
   * @param numInstances the number of instances to generate
   * @param random the random number generator
   * @param cl the cluster definition
   * @param cName the class value
   */
  private void generateGaussianExamples(
      Instances format, int numInstances, Random random, 
      SubspaceClusterDefinition cl, String cName) {

    boolean makeInteger = cl.isInteger();
    Instance example = null;
    int numAtts = m_NumAttributes;
    if (getClassFlag()) numAtts++;

    example = new Instance(numAtts);
    example.setDataset(format);
    boolean[] attributes = cl.getAttributes();
    double[] meanValue = cl.getMeanValue();
    double[] stddevValue = cl.getStddevValue();

    for (int j = 0; j < numInstances; j++) {
      int num = -1;
      for (int i = 0; i < m_NumAttributes; i++) {
        if (attributes[i]) {
          num++;
          double value = meanValue[num] + (random.nextGaussian() * stddevValue[num]);
          if (makeInteger)
            value = Math.round(value);
          example.setValue(i, value);
        } 
        else {
          example.setMissing(i);
        }
      }
      if (getClassFlag())
        example.setClassValue(cName);
      format.add(example);
    }
  }

  /**
   * Compiles documentation about the data generation after
   * the generation process
   *
   * @return string with additional information about generated dataset
   * @throws Exception no input structure has been defined
   */
  public String generateFinished() throws Exception {
    return "";
  }

  /**
   * Compiles documentation about the data generation before
   * the generation process
   *
   * @return string with additional information 
   */
  public String generateStart() {
    StringBuffer docu = new StringBuffer();

    int sumInst = 0;
    for (int cNum = 0; cNum < getClusters().length; cNum++) {
      SubspaceClusterDefinition cl  = (SubspaceClusterDefinition) getClusters()[cNum];
      docu.append("%\n");
      docu.append("% Cluster: c"+ cNum + "   ");
      switch (cl.getClusterType().getSelectedTag().getID()) {
        case UNIFORM_RANDOM: 
          docu.append("Uniform Random");
          break;
        case TOTAL_UNIFORM: 
          docu.append("Total Random");
          break;
        case GAUSSIAN: 
          docu.append("Gaussian");
          break;
      }
      if (cl.isInteger()) {
        docu.append(" / INTEGER");
      }

      docu.append("\n% ----------------------------------------------\n");
      docu.append("%"+cl.attributesToString());

      docu.append("\n% Number of Instances:            "  + cl.getInstNums() + "\n");
      docu.append(  "% Generated Number of Instances:  "  + cl.getNumInstances() + "\n");
      sumInst += cl.getNumInstances();
        }
    docu.append("%\n% ----------------------------------------------\n"); 
    docu.append("% Total Number of Instances: " + sumInst + "\n");
    docu.append("%                            in " + getClusters().length + " Cluster(s)\n%");

    return docu.toString();
  }


  /**
   * Main method for testing this class.
   *
   * @param args should contain arguments for the data producer: 
   */
  public static void main(String[] args) {
    runDataGenerator(new SubspaceCluster(), args);
  }
}