mismo.java

Java 编写的多种数据挖掘算法 包括聚类、分类、预处理等

    // Filter instances 
    if (m_Missing!=null) 
      insts = Filter.useFilter(insts, m_Missing); 

    if (m_Filter!=null)
      insts = Filter.useFilter(insts, m_Filter);     

    // convert the single-instance format to multi-instance format
    convertToMI.setInputFormat(insts);
    insts=Filter.useFilter( insts, convertToMI);

    inst = insts.instance(0);  

    if (!m_fitLogisticModels) {
      double[] result = new double[inst.numClasses()];
      for (int i = 0; i < inst.numClasses(); i++) {
        for (int j = i + 1; j < inst.numClasses(); j++) {
          if ((m_classifiers[i][j].m_alpha != null) || 
              (m_classifiers[i][j].m_sparseWeights != null)) {
            double output = m_classifiers[i][j].SVMOutput(-1, inst);
            if (output > 0) {
              result[j] += 1;
            } else {
              result[i] += 1;
            }
              }
        } 
      }
      Utils.normalize(result);
      return result;
    } else {

      // We only need to do pairwise coupling if there are more
      // then two classes.
      if (inst.numClasses() == 2) {
        double[] newInst = new double[2];
        newInst[0] = m_classifiers[0][1].SVMOutput(-1, inst);
        newInst[1] = Instance.missingValue();
        return m_classifiers[0][1].m_logistic.
          distributionForInstance(new Instance(1, newInst));
      }
      double[][] r = new double[inst.numClasses()][inst.numClasses()];
      double[][] n = new double[inst.numClasses()][inst.numClasses()];
      for (int i = 0; i < inst.numClasses(); i++) {
        for (int j = i + 1; j < inst.numClasses(); j++) {
          if ((m_classifiers[i][j].m_alpha != null) || 
              (m_classifiers[i][j].m_sparseWeights != null)) {
            double[] newInst = new double[2];
            newInst[0] = m_classifiers[i][j].SVMOutput(-1, inst);
            newInst[1] = Instance.missingValue();
            r[i][j] = m_classifiers[i][j].m_logistic.
              distributionForInstance(new Instance(1, newInst))[0];
            n[i][j] = m_classifiers[i][j].m_sumOfWeights;
              }
        }
      }
      return pairwiseCoupling(n, r);
    }
  }

  /**
   * Implements pairwise coupling.
   *
   * @param n the sum of weights used to train each model
   * @param r the probability estimate from each model
   * @return the coupled estimates
   */
  public double[] pairwiseCoupling(double[][] n, double[][] r) {

    // Initialize p and u array
    double[] p = new double[r.length];
    for (int i =0; i < p.length; i++) {
      p[i] = 1.0 / (double)p.length;
    }
    double[][] u = new double[r.length][r.length];
    for (int i = 0; i < r.length; i++) {
      for (int j = i + 1; j < r.length; j++) {
        u[i][j] = 0.5;
      }
    }

    // firstSum doesn't change
    double[] firstSum = new double[p.length];
    for (int i = 0; i < p.length; i++) {
      for (int j = i + 1; j < p.length; j++) {
        firstSum[i] += n[i][j] * r[i][j];
        firstSum[j] += n[i][j] * (1 - r[i][j]);
      }
    }

    // Iterate until convergence
    boolean changed;
    do {
      changed = false;
      double[] secondSum = new double[p.length];
      for (int i = 0; i < p.length; i++) {
        for (int j = i + 1; j < p.length; j++) {
          secondSum[i] += n[i][j] * u[i][j];
          secondSum[j] += n[i][j] * (1 - u[i][j]);
        }
      }
      for (int i = 0; i < p.length; i++) {
        if ((firstSum[i] == 0) || (secondSum[i] == 0)) {
          if (p[i] > 0) {
            changed = true;
          }
          p[i] = 0;
        } else {
          double factor = firstSum[i] / secondSum[i];
          double pOld = p[i];
          p[i] *= factor;
          if (Math.abs(pOld - p[i]) > 1.0e-3) {
            changed = true;
          }
        }
      }
      Utils.normalize(p);
      for (int i = 0; i < r.length; i++) {
        for (int j = i + 1; j < r.length; j++) {
          u[i][j] = p[i] / (p[i] + p[j]);
        }
      }
    } while (changed);
    return p;
  }

  /**
   * Returns the weights in sparse format.
   * 
   * @return the weights in sparse format
   */
  public double [][][] sparseWeights() {

    int numValues = m_classAttribute.numValues();
    double [][][] sparseWeights = new double[numValues][numValues][];

    for (int i = 0; i < numValues; i++) {
      for (int j = i + 1; j < numValues; j++) {
        sparseWeights[i][j] = m_classifiers[i][j].m_sparseWeights;
      }
    }

    return sparseWeights;
  }

  /**
   * Returns the indices in sparse format.
   * 
   * @return the indices in sparse format
   */
  public int [][][] sparseIndices() {

    int numValues = m_classAttribute.numValues();
    int [][][] sparseIndices = new int[numValues][numValues][];

    for (int i = 0; i < numValues; i++) {
      for (int j = i + 1; j < numValues; j++) {
        sparseIndices[i][j] = m_classifiers[i][j].m_sparseIndices;
      }
    }

    return sparseIndices;
  }

  /**
   * Returns the bias of each binary SMO.
   * 
   * @return the bias of each binary SMO
   */
  public double [][] bias() {

    int numValues = m_classAttribute.numValues();
    double [][] bias = new double[numValues][numValues];

    for (int i = 0; i < numValues; i++) {
      for (int j = i + 1; j < numValues; j++) {
        bias[i][j] = m_classifiers[i][j].m_b;
      }
    }

    return bias;
  }

  /**
   * Returns the number of values of the class attribute.
   * 
   * @return the number values of the class attribute
   */
  public int numClassAttributeValues() {

    return m_classAttribute.numValues();
  }

  /**
   * Returns the names of the class attributes.
   * 
   * @return the names of the class attributes
   */
  public String[] classAttributeNames() {

    int numValues = m_classAttribute.numValues();

    String[] classAttributeNames = new String[numValues];

    for (int i = 0; i < numValues; i++) {
      classAttributeNames[i] = m_classAttribute.value(i);
    }

    return classAttributeNames;
  }

  /**
   * Returns the attribute names.
   * 
   * @return the attribute names
   */
  public String[][][] attributeNames() {

    int numValues = m_classAttribute.numValues();
    String[][][] attributeNames = new String[numValues][numValues][];

    for (int i = 0; i < numValues; i++) {
      for (int j = i + 1; j < numValues; j++) {
        int numAttributes = m_classifiers[i][j].m_data.numAttributes();
        String[] attrNames = new String[numAttributes];
        for (int k = 0; k < numAttributes; k++) {
          attrNames[k] = m_classifiers[i][j].m_data.attribute(k).name();
        }
        attributeNames[i][j] = attrNames;          
      }
    }
    return attributeNames;
  }

  /**
   * Returns an enumeration describing the available options.
   *
   * @return an enumeration of all the available options.
   */
  public Enumeration listOptions() {

    Vector result = new Vector();

    Enumeration enm = super.listOptions();
    while (enm.hasMoreElements())
      result.addElement(enm.nextElement());

    result.addElement(new Option(
	"\tTurns off all checks - use with caution!\n"
	+ "\tTurning them off assumes that data is purely numeric, doesn't\n"
	+ "\tcontain any missing values, and has a nominal class. Turning them\n"
	+ "\toff also means that no header information will be stored if the\n"
	+ "\tmachine is linear. Finally, it also assumes that no instance has\n"
	+ "\ta weight equal to 0.\n"
	+ "\t(default: checks on)",
	"no-checks", 0, "-no-checks"));

    result.addElement(new Option(
          "\tThe complexity constant C. (default 1)",
          "C", 1, "-C <double>"));
    
    result.addElement(new Option(
          "\tWhether to 0=normalize/1=standardize/2=neither.\n" 
          + "\t(default 0=normalize)",
          "N", 1, "-N"));
    
    result.addElement(new Option(
          "\tUse MIminimax feature space. ",
          "I", 0, "-I"));
    
    result.addElement(new Option(
          "\tThe tolerance parameter. (default 1.0e-3)",
          "L", 1, "-L <double>"));
    
    result.addElement(new Option(
          "\tThe epsilon for round-off error. (default 1.0e-12)",
          "P", 1, "-P <double>"));
    
    result.addElement(new Option(
          "\tFit logistic models to SVM outputs. ",
          "M", 0, "-M"));
    
    result.addElement(new Option(
          "\tThe number of folds for the internal cross-validation. \n"
          + "\t(default -1, use training data)",
          "V", 1, "-V <double>"));
    
    result.addElement(new Option(
          "\tThe random number seed. (default 1)",
          "W", 1, "-W <double>"));
    
    result.addElement(new Option(
	"\tThe Kernel to use.\n"
	+ "\t(default: weka.classifiers.functions.supportVector.PolyKernel)",
	"K", 1, "-K <classname and parameters>"));

    result.addElement(new Option(
	"",
	"", 0, "\nOptions specific to kernel "
	+ getKernel().getClass().getName() + ":"));
    
    enm = ((OptionHandler) getKernel()).listOptions();
    while (enm.hasMoreElements())
      result.addElement(enm.nextElement());

    return result.elements();
  }

  /**
   * Parses a given list of options. <p/>
   * 
   <!-- options-start -->
   * Valid options are: <p/>
   * 
   * <pre> -D
   *  If set, classifier is run in debug mode and
   *  may output additional info to the console</pre>
   * 
   * <pre> -no-checks
   *  Turns off all checks - use with caution!
   *  Turning them off assumes that data is purely numeric, doesn't
   *  contain any missing values, and has a nominal class. Turning them
   *  off also means that no header information will be stored if the
   *  machine is linear. Finally, it also assumes that no instance has
   *  a weight equal to 0.
   *  (default: checks on)</pre>
   * 
   * <pre> -C &lt;double&gt;
   *  The complexity constant C. (default 1)</pre>
   * 
   * <pre> -N
   *  Whether to 0=normalize/1=standardize/2=neither.
   *  (default 0=normalize)</pre>
   * 
   * <pre> -I
   *  Use MIminimax feature space. </pre>
   * 
   * <pre> -L &lt;double&gt;
   *  The tolerance parameter. (default 1.0e-3)</pre>
   * 
   * <pre> -P &lt;double&gt;
   *  The epsilon for round-off error. (default 1.0e-12)</pre>
   * 
   * <pre> -M
   *  Fit logistic models to SVM outputs. </pre>
   * 
   * <pre> -V &lt;double&gt;
   *  The number of folds for the internal cross-validation. 
   *  (default -1, use training data)</pre>
   * 
   * <pre> -W &lt;double&gt;
   *  The random number seed. (default 1)</pre>
   * 
   * <pre> -K &lt;classname and parameters&gt;
   *  The Kernel to use.
   *  (default: weka.classifiers.functions.supportVector.PolyKernel)</pre>
   * 
   * <pre> 
   * Options specific to kernel weka.classifiers.mi.supportVector.MIPolyKernel:
   * </pre>
   * 
   * <pre> -D
   *  Enables debugging output (if available) to be printed.
   *  (default: off)</pre>
   * 
   * <pre> -no-checks
   *  Turns off all checks - use with caution!
   *  (default: checks on)</pre>
   * 
   * <pre> -C &lt;num&gt;
   *  The size of the cache (a prime number).
   *  (default: 250007)</pre>
   * 
   * <pre> -E &lt;num&gt;
   *  The Exponent to use.
   *  (default: 1.0)</pre>
   * 
   * <pre> -L
   *  Use lower-order terms.
   *  (default: no)</pre>
   * 
   <!-- options-end -->
   *
   * @param options the list of options as an array of strings
   * @throws Exception if an option is not supported 
   */
  public void setOptions(String[] options) throws Exception {
    String	tmpStr;
    String[]	tmpOptions;
    
    setChecksTurnedOff(Utils.getFlag("no-checks", options));

    tmpStr = Utils.getOption('C', options);
    if (tmpStr.length() != 0)
      setC(Double.parseDouble(tmpStr));
    else
      setC(1.0);

    tmpStr = Utils.getOption('L', options);
    if (tmpStr.length() != 0)
      setToleranceParameter(Double.parseDouble(tmpStr));
    else
      setToleranceParameter(1.0e-3);
    
    tmpStr = Utils.getOption('P', options);
    if (tmpStr.length() != 0)
      setEpsilon(new Double(tmpStr));
    else
      setEpsilon(1.0e-12);

    setMinimax(Utils.getFlag('I', options));

    tmpStr = Utils.getOption('N', options);
    if (tmpStr.length() != 0)
      setFilterType(new SelectedTag(Integer.parseInt(tmpStr), TAGS_FILTER));
    else
      setFilterType(new SelectedTag(FILTER_NORMALIZE, TAGS_FILTER));
    
    setBuildLogisticModels(Utils.getFlag('M', options));
    
    tmpStr = Utils.getOption('V', options);
    if (tmpStr.length() != 0)
      m_numFolds = Integer.parseInt(tmpStr);