costsensitiveclassifiersplitevaluator.java

Java 编写的多种数据挖掘算法 包括聚类、分类、预处理等

      ? m_AdditionalMeasures.length 
      : 0;
    Object [] resultTypes = new Object[RESULT_SIZE+addm];
    Double doub = new Double(0);
    int current = 0;
    resultTypes[current++] = doub;

    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // Timing stats
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    
    resultTypes[current++] = "";

    // add any additional measures
    for (int i=0;i<addm;i++) {
      resultTypes[current++] = doub;
    }
    if (current != RESULT_SIZE+addm) {
      throw new Error("ResultTypes didn't fit RESULT_SIZE");
    }
    return resultTypes;
  }

  /**
   * Gets the names of each of the result columns produced for a single run.
   * The number of result fields must be constant
   * for a given SplitEvaluator.
   *
   * @return an array containing the name of each result column
   */
  public String [] getResultNames() {
    int addm = (m_AdditionalMeasures != null) 
      ? m_AdditionalMeasures.length 
      : 0;
    String [] resultNames = new String[RESULT_SIZE+addm];
    int current = 0;
    resultNames[current++] = "Number_of_instances";

    // Basic performance stats - right vs wrong
    resultNames[current++] = "Number_correct";
    resultNames[current++] = "Number_incorrect";
    resultNames[current++] = "Number_unclassified";
    resultNames[current++] = "Percent_correct";
    resultNames[current++] = "Percent_incorrect";
    resultNames[current++] = "Percent_unclassified";
    resultNames[current++] = "Total_cost";
    resultNames[current++] = "Average_cost";

    // Sensitive stats - certainty of predictions
    resultNames[current++] = "Mean_absolute_error";
    resultNames[current++] = "Root_mean_squared_error";
    resultNames[current++] = "Relative_absolute_error";
    resultNames[current++] = "Root_relative_squared_error";

    // SF stats
    resultNames[current++] = "SF_prior_entropy";
    resultNames[current++] = "SF_scheme_entropy";
    resultNames[current++] = "SF_entropy_gain";
    resultNames[current++] = "SF_mean_prior_entropy";
    resultNames[current++] = "SF_mean_scheme_entropy";
    resultNames[current++] = "SF_mean_entropy_gain";

    // K&B stats
    resultNames[current++] = "KB_information";
    resultNames[current++] = "KB_mean_information";
    resultNames[current++] = "KB_relative_information";

    // Timing stats
    resultNames[current++] = "Elapsed_Time_training";
    resultNames[current++] = "Elapsed_Time_testing";
    resultNames[current++] = "UserCPU_Time_training";
    resultNames[current++] = "UserCPU_Time_testing";

    // Classifier defined extras
    resultNames[current++] = "Summary";
    // add any additional measures
    for (int i=0;i<addm;i++) {
      resultNames[current++] = m_AdditionalMeasures[i];
    }
    if (current != RESULT_SIZE+addm) {
      throw new Error("ResultNames didn't fit RESULT_SIZE");
    }
    return resultNames;
  }

  /**
   * Gets the results for the supplied train and test datasets. Now performs
   * a deep copy of the classifier before it is built and evaluated (just in case
   * the classifier is not initialized properly in buildClassifier()).
   *
   * @param train the training Instances.
   * @param test the testing Instances.
   * @return the results stored in an array. The objects stored in
   * the array may be Strings, Doubles, or null (for the missing value).
   * @throws Exception if a problem occurs while getting the results
   */
  public Object [] getResult(Instances train, Instances test)
  throws Exception {
    
    if (train.classAttribute().type() != Attribute.NOMINAL) {
      throw new Exception("Class attribute is not nominal!");
    }
    if (m_Template == null) {
      throw new Exception("No classifier has been specified");
    }
    ThreadMXBean thMonitor = ManagementFactory.getThreadMXBean();
    boolean canMeasureCPUTime = thMonitor.isThreadCpuTimeSupported();
    if(!thMonitor.isThreadCpuTimeEnabled())
      thMonitor.setThreadCpuTimeEnabled(true);
    
    int addm = (m_AdditionalMeasures != null) ? m_AdditionalMeasures.length : 0;
    Object [] result = new Object[RESULT_SIZE+addm];
    long thID = Thread.currentThread().getId();
    long CPUStartTime=-1, trainCPUTimeElapsed=-1, testCPUTimeElapsed=-1,
         trainTimeStart, trainTimeElapsed, testTimeStart, testTimeElapsed;    
    
    String costName = train.relationName() + CostMatrix.FILE_EXTENSION;
    File costFile = new File(getOnDemandDirectory(), costName);
    if (!costFile.exists()) {
      throw new Exception("On-demand cost file doesn't exist: " + costFile);
    }
    CostMatrix costMatrix = new CostMatrix(new BufferedReader(
    new FileReader(costFile)));
    
    Evaluation eval = new Evaluation(train, costMatrix);    
    m_Classifier = Classifier.makeCopy(m_Template);
    
    trainTimeStart = System.currentTimeMillis();
    if(canMeasureCPUTime)
      CPUStartTime = thMonitor.getThreadUserTime(thID);
    m_Classifier.buildClassifier(train);
    if(canMeasureCPUTime)
      trainCPUTimeElapsed = thMonitor.getThreadUserTime(thID) - CPUStartTime;
    trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
    testTimeStart = System.currentTimeMillis();
    if(canMeasureCPUTime)
      CPUStartTime = thMonitor.getThreadUserTime(thID);
    eval.evaluateModel(m_Classifier, test);
    if(canMeasureCPUTime)
      testCPUTimeElapsed = thMonitor.getThreadUserTime(thID) - CPUStartTime;
    testTimeElapsed = System.currentTimeMillis() - testTimeStart;
    thMonitor = null;
    
    m_result = eval.toSummaryString();
    // The results stored are all per instance -- can be multiplied by the
    // number of instances to get absolute numbers
    int current = 0;
    result[current++] = new Double(eval.numInstances());
    
    result[current++] = new Double(eval.correct());
    result[current++] = new Double(eval.incorrect());
    result[current++] = new Double(eval.unclassified());
    result[current++] = new Double(eval.pctCorrect());
    result[current++] = new Double(eval.pctIncorrect());
    result[current++] = new Double(eval.pctUnclassified());
    result[current++] = new Double(eval.totalCost());
    result[current++] = new Double(eval.avgCost());
    
    result[current++] = new Double(eval.meanAbsoluteError());
    result[current++] = new Double(eval.rootMeanSquaredError());
    result[current++] = new Double(eval.relativeAbsoluteError());
    result[current++] = new Double(eval.rootRelativeSquaredError());
    
    result[current++] = new Double(eval.SFPriorEntropy());
    result[current++] = new Double(eval.SFSchemeEntropy());
    result[current++] = new Double(eval.SFEntropyGain());
    result[current++] = new Double(eval.SFMeanPriorEntropy());
    result[current++] = new Double(eval.SFMeanSchemeEntropy());
    result[current++] = new Double(eval.SFMeanEntropyGain());
    
    // K&B stats
    result[current++] = new Double(eval.KBInformation());
    result[current++] = new Double(eval.KBMeanInformation());
    result[current++] = new Double(eval.KBRelativeInformation());
    
    // Timing stats
    result[current++] = new Double(trainTimeElapsed / 1000.0);
    result[current++] = new Double(testTimeElapsed / 1000.0);
    if(canMeasureCPUTime) {
      result[current++] = new Double((trainCPUTimeElapsed/1000000.0) / 1000.0);
      result[current++] = new Double((testCPUTimeElapsed /1000000.0) / 1000.0);
    }
    else {
      result[current++] = new Double(Instance.missingValue());
      result[current++] = new Double(Instance.missingValue());
    }
    
    if (m_Classifier instanceof Summarizable) {
      result[current++] = ((Summarizable)m_Classifier).toSummaryString();
    } else {
      result[current++] = null;
    }
    
    for (int i=0;i<addm;i++) {
      if (m_doesProduce[i]) {
        try {
          double dv = ((AdditionalMeasureProducer)m_Classifier).
          getMeasure(m_AdditionalMeasures[i]);
          if (!Instance.isMissingValue(dv)) {
            Double value = new Double(dv);
            result[current++] = value;
          } else {
            result[current++] = null;
          }
        } catch (Exception ex) {
          System.err.println(ex);
        }
      } else {
        result[current++] = null;
      }
    }
    
    if (current != RESULT_SIZE+addm) {
      throw new Error("Results didn't fit RESULT_SIZE");
    }
    return result;
  }

  /**
   * Returns a text description of the split evaluator.
   *
   * @return a text description of the split evaluator.
   */
  public String toString() {

    String result = "CostSensitiveClassifierSplitEvaluator: ";
    if (m_Template == null) {
      return result + "<null> classifier";
    }
    return result + m_Template.getClass().getName() + " " 
      + m_ClassifierOptions + "(version " + m_ClassifierVersion + ")";
  }
} // CostSensitiveClassifierSplitEvaluator