datadistributor.java

机器学习算法中的CHC算法

package chc;

import shared.InstanceList;
import shared.Instance;
import java.util.LinkedList;
import java.io.File;

/** Class take care of the training and testing data used for chc
  * data trials. Data 'instances' are held in InstanceLists and 
  * inducers recieve the instances from the lists.
  * 
  * Chc uses data in .name, ,.all, .train, .val, .data, and .test files
  *  .all - has all data instances for a data set.
  *  .train - has data instance used for training the innerloop
  *           inducer.
  *  .val - data for testing on the inner loop inducer.
  *  .data - trains for the outer loop inducer.
  *  .text - testing data for outer loop inducer. */
public class DataDistributor {
  /** holds the name of the data file without any extention.
    * File object was used becuase of its functions to parse
    * the name and path of the file. */
  public File file;
  /** holds the .train file data for inner loop training.  */
  private InstanceList traindata;
  /** hold the .val file data for inner loop testing. */
  private InstanceList validationdata;
  /** holds .data which is a combination of .train and .val for
    * outer loop training. */
  private InstanceList finaltrainer;
  /** holds the .test data for outer loop testing. */
  private InstanceList testdata;

  private double trainingdistribution = .6;
  private double validationdistribution = .2;
  private double testingdistribution = .2;

  private int attrselectionsize = 0;
  /** General constructor. Determines the file system and makes
    * all the InstancesLists needed for chc trials.
    * @param filename - the base name of the file with file path 
    *                   included.
    * @param system - if true the new system is used; if false the old. */
  public DataDistributor(String filename, boolean system) {
    file = new File(filename);
    if (system) {
      traindata = new InstanceList(filename,".names",".train");
      validationdata = new InstanceList(filename,".names",".val");
      testdata = new InstanceList(filename,".names",".test");
      finaltrainer = new InstanceList(filename,".names",".data");
    }
    else {
      distributedata(filename);
    }
  }

  /** distribute data is used when a single data file must be
    * broken into 3 testable parts, Training Data, Validation
    * Data, and Testing Data. If the data is already separated
    * distribute data is not needed.
    *
    * This method is technically not needed since this system is
    * never used in chc. The code is included but the method is
    * disabled.
    * @param arg - a string with the data file name root. */
  private void distributedata(String arg) {
/*    InstanceList templist = new InstanceList(arg,".names",".all");
    attrselectionsize = templist.num_attr();
    int totalinstances = templist.num_instances();
    System.out.println("    Number of Instances:  " + totalinstances);
    int traininstances = (int)(totalinstances*trainingdistribution);
    int validateinstances = (int)(totalinstances*validationdistribution);
    int testinstances = (int)(totalinstances*testingdistribution);
    
    int sumofinstances = traininstances+validateinstances+testinstances;
    if (sumofinstances < totalinstances) {
      System.out.println("Adding "+(totalinstances-sumofinstances)+" to validation data");
      validateinstances += (totalinstances-sumofinstances);
    }
    else if (sumofinstances < totalinstances) {
      System.out.println("Removing "+(sumofinstances-totalinstances)+" from training data");
      traininstances -= (sumofinstances-totalinstances);
    }
    sumofinstances = traininstances+validateinstances+testinstances;
    System.out.println(" |||||||||||||||||||||||||||||||||||||||||||||||||");
    System.out.println(" ||                      Number    Percent      ||");
    System.out.print(" || Training Data:   ");
    Hypothesis.printInt(traininstances, 10);
    System.out.print("    ");
    Hypothesis.printDouble((double)traininstances/sumofinstances, 5);
    System.out.print("       ||");
    System.out.print("\n || Validation Data: ");
    Hypothesis.printInt(validateinstances, 10);
    System.out.print("    ");
    Hypothesis.printDouble((double)validateinstances/sumofinstances, 5);
    System.out.print("       ||");
    System.out.print("\n || Testing Data:    ");
    Hypothesis.printInt(testinstances, 10);
    System.out.print("    ");
    Hypothesis.printDouble((double)testinstances/sumofinstances, 5);
    System.out.print("       ||");
    System.out.print("\n || Total:           ");
    Hypothesis.printInt(traininstances+validateinstances+testinstances, 10);
    System.out.print("                 ||");
    System.out.println("\n |||||||||||||||||||||||||||||||||||||||||||||||||");
    
    traindata = templist.split(traininstances);
    validationdata = templist.split(validateinstances);
    testdata = templist.split(testinstances);
    finaltrainer = makeFinalTrainer();
*/
  }

  /** Makes the outer loop trainer by adding all inner loop instances
    * into one InstanceList.
    *
    * Function is not used because it is used with a different file 
    * system.
    * @return - a list with the outer loop training data. */
  private InstanceList makeFinalTrainer() {
    InstanceList newlist = new InstanceList(traindata, false);
    InstanceList newlist2 = new InstanceList(validationdata, false);
    LinkedList instances = newlist2.instance_list();
    while ( !newlist2.no_instances() ) {
      newlist.add_instance((Instance)instances.removeFirst());
    }
    return newlist;
  }

  /** returns a copy of the .train data. It is copied so no changes
    * occur to the original data copy.
    * @return the list of the data. */
  public InstanceList getTrainData() { return new InstanceList(traindata, false); }

  /** returns a copy of the .val data. It is copied so no changes
    * occur to the original data copy.
    * @return the list of the data. */
  public InstanceList getValidationData() { return new InstanceList(validationdata, false); }

  /** returns a copy of the .test data. It is copied so no changes
    * occur to the original data copy.
    * @return the list of the data. */
  public InstanceList getTestData() { return new InstanceList(testdata, false); }

  /** returns a copy of the .data data. It is copied so no changes
    * occur to the original data copy.
    * @return the list of the data. */
  public InstanceList getFinalTrainer() { return new InstanceList(finaltrainer, false); }

  /** Gets the number of attribute in the data set.
    * @return - the number of attributes. */
  public int getattrselectionsize() { return testdata.num_attr(); }

}