instancesutil.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

	    revPref[1]++;
	    if (cdf2.stochasticDominatedBy(cdf1) == true) {
	      revPref[2]++;
	    }
	  }
	} 
      }
    }
    revPref[4] /= 2; 
    return revPref;
  }

  /**
   * Counts the number of redundant pairs in the sense of OLM.
   * Two instances are redundant if they are comparable and have the same
   * class value.
   * 
   * @param instances the instances to be checked
   * @return the number of redundant pairs in the given set of instances
   */
  public static int nrOfRedundant(Instances instances) {
    int n = instances.numInstances();
    int nrRedundant = 0;
    for (int i = 0; i < n; i++) {
      Instance i1 = instances.instance(i);
      for (int j = i + 1; j < n; j++) {
	Instance j1 = instances.instance(j);
	if (j1.classValue() == i1.classValue() && comparable(i1, j1) ) {
	  nrRedundant++; 
	}
      }
    }

    return nrRedundant;
  }

  /**
   * Calulates the total loss over the <code> instances </code>, 
   * using the trained <code> classifier </code> and the 
   * specified <code> lossFunction. </code>  The instances 
   * should not contain missing values in the class attribute.
   *
   * @param classifier the trained classifier to use
   * @param instances the test instances
   * @param lossFunction the loss function to use
   * @return the total loss of all the instances using the given classifier and loss function
   */
  public static double totalLoss(Classifier classifier, Instances instances, 
      NominalLossFunction lossFunction) {

    double loss = 0;
    int n = instances.numInstances();
    for (int i = 0; i < n; i++) {
      try {
	loss += lossFunction.loss(instances.instance(i).classValue(), 
	    classifier.classifyInstance(instances.instance(i)));
      } catch (Exception e) { 
	// what should we do here ?? 
      }
    }
    return loss;
  }


  /**
   * Classify a set of instances using a given classifier.  The class value
   * of the instances are set.
   *
   * @param instances the instances to be classified
   * @param classifier a built classifier 
   * @throws Exception if one of the instances could no be classified
   */
  public static void classifyInstances(Instances instances, Classifier classifier)
    throws Exception {
    
    Iterator it = new EnumerationIterator(instances.enumerateInstances());
    while(it.hasNext()) {
      Instance instance = (Instance) it.next();
      instance.setClassValue(classifier.classifyInstance(instance));
    }
  }


  /**
   * Calculates the relation (poset) formed by the instances.
   *
   * @param instances the instances for which the poset is to be formed
   * @return a <code> BooleanBitMatrix </code> for which position 
   * <code> bm.get(i,j) == true </code> iff <code> 
   * InstancesUtil.strictlySmaller(instances.instance(i), 
   * instances.instance(j)) == true </code>
   */
  public static BooleanBitMatrix getBitMatrix(Instances instances) {
    int numInstances = instances.numInstances();
    BooleanBitMatrix bm = 
      new BooleanBitMatrix(numInstances, numInstances);
    for (int i = 0; i < numInstances; i++ ) {
      Instance instance1 = instances.instance(i);
      for (int j = 0; j < numInstances; j++) {
	Instance instance2 = instances.instance(j);
	if (InstancesUtil.strictlySmaller(instance1, instance2)) {
	  bm.set(i, j); // arc from instance1 to instance2
	}
      }
    }
    return bm;
  }

  /**
   * Calculatus the number of elements in the closed interval 
   * <code> [low,up]. </code>  If the class index is set, then
   * the class attribute does not play part in the calculations,
   * this is we work in the data space.  The code also works with 
   * numeric attributes, but is primarily intended for ordinal attributes.
   *
   * @param low the lower bound of the interval
   * @param up the upper bound of the interval
   * @return the size of the interval (in floating point format)
   * @throws IllegalArgumentException if the given instances do not
   * constitute an interval.
   */
  public static double numberInInterval(Instance low, Instance up)
    throws IllegalArgumentException {
    
    Coordinates cLow = new Coordinates(low);
    Coordinates cUp = new Coordinates(up);
    if (cLow.smallerOrEqual(cUp) == false) {
      throw new IllegalArgumentException
      ("The given instances are not the bounds of an interval");
    }
    double number = 1;
    int dim = cLow.dimension();
    for (int i = 0; i < dim; i++) {
      number *= (cUp.getValue(i) - cLow.getValue(i) + 1);
    }
    return number;
  }


  /**
   * Calculatutes the number of vectors in the data space that are smaller 
   * or equal than the given instance.
   *
   * @param instance the given instance
   * @return the number of vectors in the data space smaller or equal 
   * than the given instance
   * @throws IllegalArgumentException if there are numeric attributes
   */
  public static double numberOfSmallerVectors(Instance instance) 
    throws IllegalArgumentException {
    
    double[] values = InstancesUtil.toDataDouble(instance);
    double nr = 1;

    for (int i = 0; i < values.length; i++) {
      if (instance.attribute(i).isNumeric()) {
	throw new IllegalArgumentException
	("Numeric attributes are not supported"); 
      }
      nr *= (values[i] + 1);
    }

    return nr;
  }

  /**
   * Calculatutes the number of vectors in the data space that are 
   * greater or equal than the given instance.
   *
   * @param instance the given instance
   * @return the number of vectors in the data space greater of equal
   * than the given instance
   * @throws IllegalArgumentException if there are numeric attributes
   */
  public static double numberOfGreaterVectors(Instance instance) 
    throws IllegalArgumentException {
    
    double[] values = InstancesUtil.toDataDouble(instance);
    double nr = 1;

    for (int i = 0; i < values.length; i++) {
      if (instance.attribute(i).isNumeric()) {
	throw new IllegalArgumentException
	("Numeric attributes are not supported"); 
      }
      nr *= (instance.attribute(i).numValues() - values[i]);
    }

    return nr;
  }

  /**
   * Write the instances in ARFF-format to the indicated 
   * <code> BufferedWriter </code>.
   * @param instances the instances to write
   * @param file the <code> BufferedWriter </code> to write to
   * @throws IOException if something goes wrong while writing the instances
   */
  public static void write(Instances instances, BufferedWriter file)
    throws IOException{
    
    file.write(instances.toString()); // XXX can probably be done better 
  }


  /**
   * Return a histogram of the values for the specified attribute.
   *
   * @param instances the instances
   * @param attributeIndex the attribute to consider
   * @return a <code> DiscreteEstimator </code> where the <code>i</code>th
   * @throws IllegalArgumentException if the attribute at the specified 
   * index is numeric
   */
  public static DiscreteEstimator countValues(Instances instances, int attributeIndex) 
    throws IllegalArgumentException{
    
    int numValues = instances.attribute(attributeIndex).numValues();
    if (numValues == 0) {
      throw new IllegalArgumentException
      ("Can't create histogram for numeric attribute");
    }

    DiscreteEstimator de = new DiscreteEstimator(numValues, false);
    Iterator it = new EnumerationIterator(instances.enumerateInstances());
    while (it.hasNext()) {
      Instance instance = (Instance) it.next();
      if (!instance.isMissing(attributeIndex)) {
	de.addValue(instance.value(attributeIndex), instance.weight());
      }
    }
    return de;
  }

  /**
   * Create, without replacement, a random subsample of the given size 
   * from the given instances.
   *
   * @param instances the instances to sample from
   * @param size the requested size of the sample
   * @param random the random generator to use
   * @return a sample of the requested size, drawn from the given
   * instances without replacement
   * @throws IllegalArgumentException if the size exceeds the number
   * of instances
   */
  public static Instances sampleWithoutReplacement(
      Instances instances, int size, Random random) {
    
    if (size > instances.numInstances()) {
      throw new IllegalArgumentException
      ("Size of requested sample exceeds number of instances");
    }

    int numInstances = instances.numInstances();
    int[] indices = new int[instances.numInstances()];
    for (int i = 0; i < numInstances; i++) {
      indices[i] = i;
    }

    Instances sample = new Instances(instances, size);
    int index;
    for (int i = 0; i < size; i++) {
      index = random.nextInt(numInstances--);
      sample.add(instances.instance(indices[index]));
      swap(indices, index, numInstances);
    }
    return sample;
  }

  /**
   * Swaps two elements of the given array.
   *
   * @param aa the array 
   * @param i the index of the first element
   * @param j the index of the second element
   */
  final private static void swap(int[] aa, int i, int j) {
    int tmp = aa[i];
    aa[i] = aa[j];
    aa[j] = tmp;
  }
  /**
   * Generates a random sample of instances.  Each attribute must be nominal, and the 
   * class labels are not set.
   * 
   * @param headerInfo Instances whose header information is used to determine how the 
   * set of returned instances will look
   * @param numberOfExamples the desired size of the returned set
   * @param random the random number generator to use
   * @return a set of Instances containing the random sample.
   * @throws IllegalArgumentException if numeric attributes are given
   */
  public static Instances generateRandomSample(
      Instances headerInfo, int numberOfExamples, Random random) 
    throws IllegalArgumentException {
    
    int n = headerInfo.numAttributes();
    double[] info = new double[n];
    int classIndex = headerInfo.classIndex();
    for (int i = 0; i < n; i++) {
      info[i] = headerInfo.attribute(i).numValues();
      if (i != classIndex && info[i] == 0) {
	throw new IllegalArgumentException
	("Numeric attributes are currently not supported");
      }
    }
    Instances sample = new Instances(headerInfo, numberOfExamples);
    sample.setRelationName(headerInfo.relationName() + 
	".random.sample.of." + numberOfExamples);
    for (int i = 0; i < numberOfExamples; i++) {
      sample.add(randomSample(info, classIndex, random));
    }
    return sample;
  }
  /**
   * Generates a random instance.
   * 
   * @param info array that gives for each attribute the number of possible values
   * @param classIndex the index of the class attribute
   * @param random the random number generator used
   * @return a random instance
   */
  private static Instance randomSample(double[] info, 
      int classIndex, Random random) {
    
    double[] attValues = new double[info.length];
    for (int i = 0; i < attValues.length; i++) {
      if (i != classIndex) {
	attValues[i] = random.nextInt( (int) info[i]); 
      }
    }
    return new Instance(1, attValues);
  }



  /**
   * Returns an array containing the attribute values (in internal floating 
   * point format) of the given instance in data space, this is, the class 
   * attribute (if any) is removed.
   *
   * @param instance the instance to get the attribute values from
   * @return array of doubles containing the attribute values
   */
  public static double[] toDataDouble(Instance instance) {
    double[] vector = null;
    int classIndex = instance.classIndex();
    if(classIndex >= 0) {
      vector = new double[instance.numAttributes() - 1];
    } else {
      vector = new double[instance.numAttributes()];
    }
    int index = 0;
    for (int i = 0; i < instance.numAttributes(); i++) {
      if(i != classIndex) {
	vector[index++] = instance.value(i);
      }
    }
    return vector;
  }

  /**
   * Computes the minimal extension for a given instance.
   *
   * @param instances the set of instances 
   * @param instance the instance for which the minimal extension is to be
   * calculated
   * @return the value of the minimal extension, in internal floating point
   * format
   */
  public static double minimalExtension(Instances instances, Instance instance) {
    return minimalExtension(instances, instance, 0);
  }

  /**
   * Computes the minimal extension of a given instance, but the 
   * minimal value returned is <code> minValue. </code>  This method
   * may have its applications when the training set is divided into
   * multiple Instances objects.
   *
   * @param instances the set of instances
   * @param instance the instance for which the minimal extension is to 
   * be calculated
   * @param minValue a double indicating the minimal value that should 
   * be returned
   * @return the label of the minimal extension, in internal floating point format
   */
  public static double minimalExtension(
      Instances instances, Instance instance, double minValue) {
    
    double value = minValue;

    Iterator it = 
      new EnumerationIterator(instances.enumerateInstances());
    while(it.hasNext()) {
      Instance tmp = (Instance) it.next();
      if (tmp.classValue() > value 
	  && InstancesUtil.smallerOrEqual(tmp, instance) ) {
	value = tmp.classValue();
      }
    }
    return value;
  }

  /**
   * Computes the maximal extension for a given instance.
   *
   * @param instances the set of instances 
   * @param instance the instance for which the minimal extension is to be
   * calculated
   * @return the value of the minimal extension, in internal floating point
   * format
   */
  public static double maximalExtension(Instances instances, Instance instance) {
    return maximalExtension(instances, instance, instances.numClasses() - 1);
  }

  /**
   * Computes the maximal extension of a given instance, but the 
   * maximal value returned is <code> maxValue. </code>  This method
   * may have its applications when the training set is divided into
   * multiple Instances objects.
   *
   * @param instances the set of instances
   * @param instance the instance for which the maximal extension is to 
   * be calculated
   * @param maxValue a double indicating the maximal value that should 
   * be returned
   * @return the value of the minimal extension, in internal floating point
   * format
   */
  public static double maximalExtension(
      Instances instances, Instance instance, double maxValue) {
    
    double value = maxValue;

    Iterator it = 
      new EnumerationIterator(instances.enumerateInstances());
    while(it.hasNext()) {
      Instance tmp = (Instance) it.next();
      if (tmp.classValue() < value 
	  && InstancesUtil.smallerOrEqual(instance, tmp) ) {
	value = tmp.classValue();
      }
    }
    return value;
  }
}