instances.java

:<<数据挖掘--实用机器学习技术及java实现>>一书的配套源程序

	       (tokenizer.ttype == '"')) {
      tokenizer.ttype = StreamTokenizer.TT_WORD;
    } else if ((tokenizer.ttype == StreamTokenizer.TT_WORD) &&
	       (tokenizer.sval.equals("?"))){
      tokenizer.ttype = '?';
    }
  }
	
  /**
   * Initializes the StreamTokenizer used for reading the ARFF file.
   *
   * @param tokenizer the stream tokenizer
   */
  private void initTokenizer(StreamTokenizer tokenizer){

    tokenizer.resetSyntax();         
    tokenizer.whitespaceChars(0, ' ');    
    tokenizer.wordChars(' '+1,'\u00FF');
    tokenizer.whitespaceChars(',',',');
    tokenizer.commentChar('%');
    tokenizer.quoteChar('"');
    tokenizer.quoteChar('\'');
    tokenizer.ordinaryChar('{');
    tokenizer.ordinaryChar('}');
    tokenizer.eolIsSignificant(true);
  }
 
  /**
   * Returns string including all instances, their weights and
   * their indices in the original dataset.
   *
   * @return description of instance and its weight as a string
   */
  private String instancesAndWeights(){

    StringBuffer text = new StringBuffer();

    for (int i = 0; i < numInstances(); i++) {
      text.append(instance(i) + " " + instance(i).weight());
      if (i < numInstances() - 1) {
	text.append("\n");
      }
    }
    return text.toString();
  }
  
  /**
   * Implements quicksort.
   *
   * @param attIndex the attribute's index
   * @param lo0 the first index of the subset to be sorted
   * @param hi0 the last index of the subset to be sorted
   */
  private void quickSort(int attIndex, int lo0, int hi0) {
    
    int lo = lo0, hi = hi0;
    double mid, midPlus, midMinus;
    
    if (hi0 > lo0) {
      
      // Arbitrarily establishing partition element as the 
      // midpoint of the array.
      mid = instance((lo0 + hi0) / 2).value(attIndex);
      midPlus = mid + 1e-6;
      midMinus = mid - 1e-6;

      // loop through the array until indices cross
      while(lo <= hi) {
	
	// find the first element that is greater than or equal to 
	// the partition element starting from the left Index.
	while ((instance(lo).value(attIndex) < 
		midMinus) && (lo < hi0)) {
	  ++lo;
	}
	
	// find an element that is smaller than or equal to
	// the partition element starting from the right Index.
	while ((instance(hi).value(attIndex)  > 
		midPlus) && (hi > lo0)) {
	  --hi;
	}
	
	// if the indexes have not crossed, swap
	if(lo <= hi) {
	  swap(lo,hi);
	  ++lo;
	  --hi;
	}
      }
      
      // If the right index has not reached the left side of array
      // must now sort the left partition.
      if(lo0 < hi) {
	quickSort(attIndex,lo0,hi);
      }
      
      // If the left index has not reached the right side of array
      // must now sort the right partition.
      if(lo < hi0) {
	quickSort(attIndex,lo,hi0);
      }
    }
  }

  /**
   * Reads and skips all tokens before next end of line token.
   *
   * @param tokenizer the stream tokenizer
   */
  private void readTillEOL(StreamTokenizer tokenizer) 
       throws IOException{
    
    while (tokenizer.nextToken() != StreamTokenizer.TT_EOL) {};
    tokenizer.pushBack();
  }

  /**
   * Help function needed for stratification of set.
   *
   * @param numFolds the number of folds for the stratification
   */
  private void stratStep (int numFolds){
    
    FastVector newVec = new FastVector(m_Instances.capacity());
    int start = 0, j;

    // create stratified batch
    while (newVec.size() < numInstances()) {
      j = start;
      while (j < numInstances()) {
	newVec.addElement(instance(j));
	j = j + numFolds;
      }
      start++;
    }
    m_Instances = newVec;
  }
  
  /**
   * Swaps two instances in the set.
   *
   * @param i the first instance's index
   * @param j the second instance's index
   */
  private void swap(int i, int j){
    
    m_Instances.swap(i, j);
  }

  /**
   * Merges two sets of Instances together. The resulting set will have
   * all the attributes of the first set plus all the attributes of the 
   * second set. The number of instances in both sets must be the same.
   *
   * @param first the first set of Instances
   * @param second the second set of Instances
   * @return the merged set of Instances
   * @exception IllegalArgumentException if the datasets are not the same size
   */
  public static Instances mergeInstances(Instances first, Instances second) {

    if (first.numInstances() != second.numInstances()) {
      throw new IllegalArgumentException("Instance sets must be of the same size");
    }

    // Create the vector of merged attributes
    FastVector newAttributes = new FastVector();
    for (int i = 0; i < first.numAttributes(); i++) {
      newAttributes.addElement(first.attribute(i));
    }
    for (int i = 0; i < second.numAttributes(); i++) {
      newAttributes.addElement(second.attribute(i));
    }
    
    // Create the set of Instances
    Instances merged = new Instances(first.relationName() + '_'
				     + second.relationName(), 
				     newAttributes, 
				     first.numInstances());
    // Merge each instance
    for (int i = 0; i < first.numInstances(); i++) {
      merged.add(first.instance(i).mergeInstance(second.instance(i)));
    }
    return merged;
  }

  /**
   * Method for testing this class.
   *
   * @param argv should contain one element: the name of an ARFF file
   */
  public static void test(String [] argv) {

    Instances instances, secondInstances, train, test, transformed, empty;
    Instance instance;
    Random random = new Random(2);
    Reader reader;
    int start, num;
    double newWeight;
    FastVector testAtts, testVals;
    int i,j;
    
    try{
      if (argv.length > 1) {
	throw (new Exception("Usage: Instances [<filename>]"));
      }
      
      // Creating set of instances from scratch
      testVals = new FastVector(2);
      testVals.addElement("first_value");
      testVals.addElement("second_value");
      testAtts = new FastVector(2);
      testAtts.addElement(new Attribute("nominal_attribute", testVals));
      testAtts.addElement(new Attribute("numeric_attribute"));
      instances = new Instances("test_set", testAtts, 10);
      instances.add(new Instance(instances.numAttributes()));
      instances.add(new Instance(instances.numAttributes()));
      instances.add(new Instance(instances.numAttributes()));
      instances.setClassIndex(0);
      System.out.println("\nSet of instances created from scratch:\n");
      System.out.println(instances);
      
      if (argv.length == 1) {
	String filename = argv[0];
	reader = new FileReader(filename);
	
	// Read first five instances and print them
	System.out.println("\nFirst five instances from file:\n");
	instances = new Instances(reader, 1);
	instances.setClassIndex(instances.numAttributes() - 1);
	i = 0;
	while ((i < 5) && (instances.readInstance(reader))) {
	  i++;
	}
	System.out.println(instances);

	// Read all the instances in the file
	reader = new FileReader(filename);
	instances = new Instances(reader);

	// Make the last attribute be the class 
	instances.setClassIndex(instances.numAttributes() - 1);
	
	// Print header and instances.
	System.out.println("\nDataset:\n");
	System.out.println(instances);
	System.out.println("\nClass index: "+instances.classIndex());
      }
      
      // Test basic methods based on class index.
      System.out.println("\nClass name: "+instances.classAttribute().name());
      System.out.println("\nClass index: "+instances.classIndex());
      System.out.println("\nClass is nominal: " +
			 instances.classAttribute().isNominal());
      System.out.println("\nClass is numeric: " +
			 instances.classAttribute().isNumeric());
      System.out.println("\nClasses:\n");
      for (i = 0; i < instances.numClasses(); i++) {
	System.out.println(instances.classAttribute().value(i));
      }
      System.out.println("\nClass values and labels of instances:\n");
      for (i = 0; i < instances.numInstances(); i++) {
	Instance inst = instances.instance(i);
	System.out.print(inst.classValue() + "\t");
	System.out.print(inst.toString(inst.classIndex()));
	if (instances.instance(i).classIsMissing()) {
	  System.out.println("\tis missing");
	} else {
	  System.out.println();
	}
      }
      
      // Create random weights.
      System.out.println("\nCreating random weights for instances.");
      for (i = 0; i < instances.numInstances(); i++) {
	instances.instance(i).setWeight(random.nextDouble()); 
      }
      
      // Print all instances and their weights (and the sum of weights).
      System.out.println("\nInstances and their weights:\n");
      System.out.println(instances.instancesAndWeights());
      System.out.print("\nSum of weights: ");
      System.out.println(instances.sumOfWeights());
      
      // Insert an attribute
      secondInstances = new Instances(instances);
      Attribute testAtt = new Attribute("Inserted");
      secondInstances.insertAttributeAt(testAtt, 0);
      System.out.println("\nSet with inserted attribute:\n");
      System.out.println(secondInstances);
      System.out.println("\nClass name: "
			 + secondInstances.classAttribute().name());
      
      // Delete the attribute
      secondInstances.deleteAttributeAt(0);
      System.out.println("\nSet with attribute deleted:\n");
      System.out.println(secondInstances);
      System.out.println("\nClass name: "
			 + secondInstances.classAttribute().name());
      
      // Test if headers are equal
      System.out.println("\nHeaders equal: "+
			 instances.equalHeaders(secondInstances) + "\n");
      
      // Print data in internal format.
      System.out.println("\nData (internal values):\n");
      for (i = 0; i < instances.numInstances(); i++) {
	for (j = 0; j < instances.numAttributes(); j++) {
	  if (instances.instance(i).isMissing(j)) {
	    System.out.print("? ");
	  } else {
	    System.out.print(instances.instance(i).value(j) + " ");
	  }
	}
	System.out.println();
      }
      
      // Just print header
      System.out.println("\nEmpty dataset:\n");
      empty = new Instances(instances, 0);
      System.out.println(empty);
      System.out.println("\nClass name: "+empty.classAttribute().name());

      // Create copy and rename an attribute and a value (if possible)
      if (empty.classAttribute().isNominal()) {
	Instances copy = new Instances(empty, 0);
	copy.renameAttribute(copy.classAttribute(), "new_name");
	copy.renameAttributeValue(copy.classAttribute(), 
				  copy.classAttribute().value(0), 
				  "new_val_name");
	System.out.println("\nDataset with names changed:\n" + copy);
	System.out.println("\nOriginal dataset:\n" + empty);
      }

      // Create and prints subset of instances.
      start = instances.numInstances() / 4;
      num = instances.numInstances() / 2;
      System.out.print("\nSubset of dataset: ");
      System.out.println(num + " instances from " + (start + 1) 
			 + ". instance");
      secondInstances = new Instances(instances, start, num);
      System.out.println("\nClass name: "
			 + secondInstances.classAttribute().name());

      // Print all instances and their weights (and the sum of weights).
      System.out.println("\nInstances and their weights:\n");
      System.out.println(secondInstances.instancesAndWeights());
      System.out.print("\nSum of weights: ");
      System.out.println(secondInstances.sumOfWeights());
      
      // Create and print training and test sets for 3-fold
      // cross-validation.
      System.out.println("\nTrain and test folds for 3-fold CV:");
      if (instances.classAttribute().isNominal()) {
	instances.stratify(3);
      }
      for (j = 0; j < 3; j++) {
        train = instances.trainCV(3,j);
	test = instances.testCV(3,j);
                      
	// Print all instances and their weights (and the sum of weights).
	System.out.println("\nTrain: ");
	System.out.println("\nInstances and their weights:\n");
	System.out.println(train.instancesAndWeights());
	System.out.print("\nSum of weights: ");
	System.out.println(train.sumOfWeights());
	System.out.println("\nClass name: "+train.classAttribute().name());
	System.out.println("\nTest: ");
	System.out.println("\nInstances and their weights:\n");
	System.out.println(test.instancesAndWeights());
	System.out.print("\nSum of weights: ");
	System.out.println(test.sumOfWeights());
	System.out.println("\nClass name: "+test.classAttribute().name());
      }

      // Randomize instances and print them.
      System.out.println("\nRandomized dataset:");
      instances.randomize(random);
      
      // Print all instances and their weights (and the sum of weights).
      System.out.println("\nInstances and their weights:\n");
      System.out.println(instances.instancesAndWeights());
      System.out.print("\nSum of weights: ");
      System.out.println(instances.sumOfWeights());

      // Sort instances according to first attribute and
      // print them.
      System.out.print("\nInstances sorted according to first attribute:\n ");
      instances.sort(0);
        
      // Print all instances and their weights (and the sum of weights).
      System.out.println("\nInstances and their weights:\n");
      System.out.println(instances.instancesAndWeights());
      System.out.print("\nSum of weights: ");
      System.out.println(instances.sumOfWeights());
    } catch (Exception e) {
      e.printStackTrace(); 
    }
  }

  /**
   * Main method for this class -- just prints a summary of a set
   * of instances.
   *
   * @param argv should contain one element: the name of an ARFF file
   */
  public static void main(String [] args) {

    try {
      Reader r = null;
      if (args.length > 1) {
	throw (new Exception("Usage: Instances <filename>"));
      } else if (args.length == 0) {
        r = new BufferedReader(new InputStreamReader(System.in));
      } else {
        r = new BufferedReader(new FileReader(args[0]));
      }
      Instances i = new Instances(r);
      System.out.println(i.toSummaryString());
    } catch (Exception ex) {
      System.err.println(ex.getMessage());
    }
  }
}