evaluation.java

一个数据挖掘软件ALPHAMINERR的整个过程的JAVA版源代码

   * @param instance the test instance to be classified
   * @exception Exception if model could not be evaluated 
   * successfully
   */
  public void evaluateModelOnce(double prediction,
				Instance instance) throws Exception {
    
    if (m_ClassIsNominal) {
      updateStatsForClassifier(makeDistribution(prediction), 
			       instance);
    } else {
      updateStatsForPredictor(prediction, instance);
    }
  }


  /**
   * Wraps a static classifier in enough source to test using the weka
   * class libraries.
   *
   * @param classifier a Sourcable Classifier
   * @param className the name to give to the source code class
   * @return the source for a static classifier that can be tested with
   * weka libraries.
   */
  protected static String wekaStaticWrapper(Sourcable classifier, 
                                            String className) 
    throws Exception {
    
    //String className = "StaticClassifier";
    String staticClassifier = classifier.toSource(className);
    return "package weka.classifiers;\n"
    +"import weka.core.Attribute;\n"
    +"import weka.core.Instance;\n"
    +"import weka.core.Instances;\n"
    +"import weka.classifiers.Classifier;\n\n"
    +"public class WekaWrapper extends Classifier {\n\n"
    +"  public void buildClassifier(Instances i) throws Exception {\n"
    +"  }\n\n"
    +"  public double classifyInstance(Instance i) throws Exception {\n\n"
    +"    Object [] s = new Object [i.numAttributes()];\n"
    +"    for (int j = 0; j < s.length; j++) {\n"
    +"      if (!i.isMissing(j)) {\n"
    +"        if (i.attribute(j).type() == Attribute.NOMINAL) {\n"
    +"          s[j] = i.attribute(j).value((int) i.value(j));\n"
    +"        } else if (i.attribute(j).type() == Attribute.NUMERIC) {\n"
    +"          s[j] = new Double(i.value(j));\n"
    +"        }\n"
    +"      }\n"
    +"    }\n"
    +"    return " + className + ".classify(s);\n"
    +"  }\n\n"
    +"}\n\n"
    +staticClassifier; // The static classifer class
  }

  /**
   * Gets the number of test instances that had a known class value
   * (actually the sum of the weights of test instances with known 
   * class value).
   *
   * @return the number of test instances with known class
   */
  public final double numInstances() {
    
    return m_WithClass;
  }

  /**
   * Gets the number of instances incorrectly classified (that is, for
   * which an incorrect prediction was made). (Actually the sum of the weights
   * of these instances)
   *
   * @return the number of incorrectly classified instances 
   */
  public final double incorrect() {

    return m_Incorrect;
  }

  /**
   * Gets the percentage of instances incorrectly classified (that is, for
   * which an incorrect prediction was made).
   *
   * @return the percent of incorrectly classified instances 
   * (between 0 and 100)
   */
  public final double pctIncorrect() {

    return 100 * m_Incorrect / m_WithClass;
  }

  /**
   * Gets the total cost, that is, the cost of each prediction times the
   * weight of the instance, summed over all instances.
   *
   * @return the total cost
   */
  public final double totalCost() {

    return m_TotalCost;
  }
  
  /**
   * Gets the average cost, that is, total cost of misclassifications
   * (incorrect plus unclassified) over the total number of instances.
   *
   * @return the average cost.  
   */
  public final double avgCost() {

    return m_TotalCost / m_WithClass;
  }

  /**
   * Gets the number of instances correctly classified (that is, for
   * which a correct prediction was made). (Actually the sum of the weights
   * of these instances)
   *
   * @return the number of correctly classified instances
   */
  public final double correct() {
    
    return m_Correct;
  }

  /**
   * Gets the percentage of instances correctly classified (that is, for
   * which a correct prediction was made).
   *
   * @return the percent of correctly classified instances (between 0 and 100)
   */
  public final double pctCorrect() {
    
    return 100 * m_Correct / m_WithClass;
  }
  
  /**
   * Gets the number of instances not classified (that is, for
   * which no prediction was made by the classifier). (Actually the sum
   * of the weights of these instances)
   *
   * @return the number of unclassified instances
   */
  public final double unclassified() {
    
    return m_Unclassified;
  }

  /**
   * Gets the percentage of instances not classified (that is, for
   * which no prediction was made by the classifier).
   *
   * @return the percent of unclassified instances (between 0 and 100)
   */
  public final double pctUnclassified() {
    
    return 100 * m_Unclassified / m_WithClass;
  }

  /**
   * Returns the estimated error rate or the root mean squared error
   * (if the class is numeric). If a cost matrix was given this
   * error rate gives the average cost.
   *
   * @return the estimated error rate (between 0 and 1, or between 0 and 
   * maximum cost)
   */
  public final double errorRate() {

    if (!m_ClassIsNominal) {
      return Math.sqrt(m_SumSqrErr / m_WithClass);
    }
    if (m_CostMatrix == null) {
      return m_Incorrect / m_WithClass;
    } else {
      return avgCost();
    }
  }

  /**
   * Returns value of kappa statistic if class is nominal.
   *
   * @return the value of the kappa statistic
   */
  public final double kappa() {
    

    double[] sumRows = new double[m_ConfusionMatrix.length];
    double[] sumColumns = new double[m_ConfusionMatrix.length];
    double sumOfWeights = 0;
    for (int i = 0; i < m_ConfusionMatrix.length; i++) {
      for (int j = 0; j < m_ConfusionMatrix.length; j++) {
	sumRows[i] += m_ConfusionMatrix[i][j];
	sumColumns[j] += m_ConfusionMatrix[i][j];
	sumOfWeights += m_ConfusionMatrix[i][j];
      }
    }
    double correct = 0, chanceAgreement = 0;
    for (int i = 0; i < m_ConfusionMatrix.length; i++) {
      chanceAgreement += (sumRows[i] * sumColumns[i]);
      correct += m_ConfusionMatrix[i][i];
    }
    chanceAgreement /= (sumOfWeights * sumOfWeights);
    correct /= sumOfWeights;

    if (chanceAgreement < 1) {
      return (correct - chanceAgreement) / (1 - chanceAgreement);
    } else {
      return 1;
    }
  }

  /**
   * Returns the correlation coefficient if the class is numeric.
   *
   * @return the correlation coefficient
   * @exception Exception if class is not numeric
   */
  public final double correlationCoefficient() throws Exception {

    if (m_ClassIsNominal) {
      throw
	new Exception("Can't compute correlation coefficient: " + 
		      "class is nominal!");
    }

    double correlation = 0;
    double varActual = 
      m_SumSqrClass - m_SumClass * m_SumClass / m_WithClass;
    double varPredicted = 
      m_SumSqrPredicted - m_SumPredicted * m_SumPredicted / 
      m_WithClass;
    double varProd = 
      m_SumClassPredicted - m_SumClass * m_SumPredicted / m_WithClass;

    if (Utils.smOrEq(varActual * varPredicted, 0.0)) {
      correlation = 0.0;
    } else {
      correlation = varProd / Math.sqrt(varActual * varPredicted);
    }

    return correlation;
  }

  /**
   * Returns the mean absolute error. Refers to the error of the
   * predicted values for numeric classes, and the error of the 
   * predicted probability distribution for nominal classes.
   *
   * @return the mean absolute error 
   */
  public final double meanAbsoluteError() {

    return m_SumAbsErr / m_WithClass;
  }

  /**
   * Returns the mean absolute error of the prior.
   *
   * @return the mean absolute error 
   */
  public final double meanPriorAbsoluteError() {

    return m_SumPriorAbsErr / m_WithClass;
  }

  /**
   * Returns the relative absolute error.
   *
   * @return the relative absolute error 
   * @exception Exception if it can't be computed
   */
  public final double relativeAbsoluteError() throws Exception {

    return 100 * meanAbsoluteError() / meanPriorAbsoluteError();
  }
  
  /**
   * Returns the root mean squared error.
   *
   * @return the root mean squared error 
   */
  public final double rootMeanSquaredError() {

    return Math.sqrt(m_SumSqrErr / m_WithClass);
  }
  
  /**
   * Returns the root mean prior squared error.
   *
   * @return the root mean prior squared error 
   */
  public final double rootMeanPriorSquaredError() {

    return Math.sqrt(m_SumPriorSqrErr / m_WithClass);
  }
  
  /**
   * Returns the root relative squared error if the class is numeric.
   *
   * @return the root relative squared error 
   */
  public final double rootRelativeSquaredError() {

    return 100.0 * rootMeanSquaredError() / 
      rootMeanPriorSquaredError();
  }

  /**
   * Calculate the entropy of the prior distribution
   *
   * @return the entropy of the prior distribution
   * @exception Exception if the class is not nominal
   */
  public final double priorEntropy() throws Exception {

    if (!m_ClassIsNominal) {
      throw
	new Exception("Can't compute entropy of class prior: " + 
		      "class numeric!");
    }

    double entropy = 0;
    for(int i = 0; i < m_NumClasses; i++) {
      entropy -= m_ClassPriors[i] / m_ClassPriorsSum 
	* Utils.log2(m_ClassPriors[i] / m_ClassPriorsSum);
    }
    return entropy;
  }


  /**
   * Return the total Kononenko & Bratko Information score in bits
   *
   * @return the K&B information score
   * @exception Exception if the class is not nominal
   */
  public final double KBInformation() throws Exception {

    if (!m_ClassIsNominal) {
      throw
	new Exception("Can't compute K&B Info score: " + 
		      "class numeric!");
    }
    return m_SumKBInfo;
  }

  /**
   * Return the Kononenko & Bratko Information score in bits per 
   * instance.
   *
   * @return the K&B information score
   * @exception Exception if the class is not nominal
   */
  public final double KBMeanInformation() throws Exception {

    if (!m_ClassIsNominal) {
      throw
	new Exception("Can't compute K&B Info score: "
		       + "class numeric!");
    }
    return m_SumKBInfo / m_WithClass;
  }

  /**
   * Return the Kononenko & Bratko Relative Information score
   *
   * @return the K&B relative information score
   * @exception Exception if the class is not nominal
   */
  public final double KBRelativeInformation() throws Exception {

    if (!m_ClassIsNominal) {
      throw
	new Exception("Can't compute K&B Info score: " + 
		      "class numeric!");
    }
    return 100.0 * KBInformation() / priorEntropy();
  }

  /**
   * Returns the total entropy for the null model
   * 
   * @return the total null model entropy
   */
  public final double SFPriorEntropy() {

    return m_SumPriorEntropy;
  }

  /**
   * Returns the entropy per instance for the null model
   * 
   * @return the null model entropy per instance
   */
  public final double SFMeanPriorEntropy() {

    return m_SumPriorEntropy / m_WithClass;
  }

  /**
   * Returns the total entropy for the scheme
   * 
   * @return the total scheme entropy