randomsearch.java

一个数据挖掘软件ALPHAMINERR的整个过程的JAVA版源代码

    if (m_verbose) {
      options[current++] = "-V";
    }

    if (!(getStartSet().equals(""))) {
      options[current++] = "-P";
      options[current++] = ""+startSetToString();
    }

    options[current++] = "-F";
    options[current++] = "" + m_searchSize;

    while (current < options.length) {
      options[current++] = "";
    }

    return  options;
  }

  /**
   * converts the array of starting attributes to a string. This is
   * used by getOptions to return the actual attributes specified
   * as the starting set. This is better than using m_startRanges.getRanges()
   * as the same start set can be specified in different ways from the
   * command line---eg 1,2,3 == 1-3. This is to ensure that stuff that
   * is stored in a database is comparable.
   * @return a comma seperated list of individual attribute numbers as a String
   */
  private String startSetToString() {
    StringBuffer FString = new StringBuffer();
    boolean didPrint;
    
    if (m_starting == null) {
      return getStartSet();
    }

    for (int i = 0; i < m_starting.length; i++) {
      didPrint = false;
      
      if ((m_hasClass == false) || 
	  (m_hasClass == true && i != m_classIndex)) {
	FString.append((m_starting[i] + 1));
	didPrint = true;
      }
      
      if (i == (m_starting.length - 1)) {
	FString.append("");
      }
      else {
	if (didPrint) {
	  FString.append(",");
	  }
      }
    }

    return FString.toString();
  }

  /**
   * prints a description of the search
   * @return a description of the search as a string
   */
  public String toString() {
    StringBuffer text = new StringBuffer();
    
    text.append("\tRandom search.\n\tStart set: ");
    if (m_starting == null) {
      text.append("no attributes\n");
    }
    else {
      text.append(startSetToString()+"\n");
    }
    text.append("\tNumber of iterations: "+m_iterations+" ("
		+(m_searchSize * 100.0)+"% of the search space)\n");
    text.append("\tMerit of best subset found: "
		+Utils.doubleToString(Math.abs(m_bestMerit),8,3)+"\n");

    return text.toString();
  }

  /**
   * Searches the attribute subset space randomly.
   *
   * @param ASEvaluator the attribute evaluator to guide the search
   * @param data the training instances.
   * @return an array (not necessarily ordered) of selected attribute indexes
   * @exception Exception if the search can't be completed
   */
   public int[] search (ASEvaluation ASEval, Instances data)
     throws Exception {
     double best_merit;
     int sizeOfBest = m_numAttribs;
     BitSet temp;
     m_bestGroup = new BitSet(m_numAttribs);
     
     m_onlyConsiderBetterAndSmaller = false;
     if (!(ASEval instanceof SubsetEvaluator)) {
       throw  new Exception(ASEval.getClass().getName() 
			    + " is not a " 
			    + "Subset evaluator!");
     }

     m_random = new Random(m_seed);
     
     if (ASEval instanceof UnsupervisedSubsetEvaluator) {
       m_hasClass = false;
     }
     else {
       m_hasClass = true;
       m_classIndex = data.classIndex();
     }
     
     SubsetEvaluator ASEvaluator = (SubsetEvaluator)ASEval;
     m_numAttribs = data.numAttributes();

     m_startRange.setUpper(m_numAttribs-1);
     if (!(getStartSet().equals(""))) {
       m_starting = m_startRange.getSelection();
     }

     // If a starting subset has been supplied, then initialise the bitset
     if (m_starting != null) {
       for (int i = 0; i < m_starting.length; i++) {
	 if ((m_starting[i]) != m_classIndex) {
	   m_bestGroup.set(m_starting[i]);
	 }
       }
       m_onlyConsiderBetterAndSmaller = true;
       best_merit = ASEvaluator.evaluateSubset(m_bestGroup);
       sizeOfBest = countFeatures(m_bestGroup);
     } else {
       // do initial random subset
       m_bestGroup = generateRandomSubset();
       best_merit = ASEvaluator.evaluateSubset(m_bestGroup);
     }
     
     if (m_verbose) {
       System.out.println("Initial subset ("
			  +Utils.doubleToString(Math.
						abs(best_merit),8,5)
			  +"): "+printSubset(m_bestGroup));
     }

     int i;
     if (m_hasClass) {
       i = m_numAttribs -1;
     } else {
       i = m_numAttribs;
     }
     m_iterations = (int)((m_searchSize * Math.pow(2, i)));
     
     int tempSize;
     double tempMerit;
     // main loop
     for (i=0;i<m_iterations;i++) {
       temp = generateRandomSubset();
       if (m_onlyConsiderBetterAndSmaller) {
	 tempSize = countFeatures(temp);
	 if (tempSize <= sizeOfBest) {
	   tempMerit = ASEvaluator.evaluateSubset(temp);
	   if (tempMerit >= best_merit) {
	     sizeOfBest = tempSize;
	     m_bestGroup = temp;
	     best_merit = tempMerit;
	     if (m_verbose) {
	       System.out.print("New best subset ("
				  +Utils.doubleToString(Math.
							abs(best_merit),8,5)
				  +"): "+printSubset(m_bestGroup) + " :");
	       System.out.println(Utils.
				  doubleToString((((double)i)/
						  ((double)m_iterations)*
						  100.0),5,1)
				  +"% done");
	     }
	   }
	 }
       } else {
	 tempMerit = ASEvaluator.evaluateSubset(temp);
	 if (tempMerit > best_merit) {
	   m_bestGroup = temp;
	   best_merit = tempMerit;
	   if (m_verbose) {
	     System.out.print("New best subset ("
				+Utils.doubleToString(Math.abs(best_merit),8,5)
				+"): "+printSubset(m_bestGroup) + " :");
	     System.out.println(Utils.
				doubleToString((((double)i)/
						((double)m_iterations)
						*100.0),5,1)
				+"% done");
	   }
	 }
       }
     }
     m_bestMerit = best_merit;
     return attributeList(m_bestGroup);
   }

  /**
   * prints a subset as a series of attribute numbers
   * @param temp the subset to print
   * @return a subset as a String of attribute numbers
   */
  private String printSubset(BitSet temp) {
    StringBuffer text = new StringBuffer();

    for (int j=0;j<m_numAttribs;j++) {
      if (temp.get(j)) {
        text.append((j+1)+" ");
      }
    }
    return text.toString();
  }

  /**
   * converts a BitSet into a list of attribute indexes 
   * @param group the BitSet to convert
   * @return an array of attribute indexes
   **/
  private int[] attributeList (BitSet group) {
    int count = 0;
    
    // count how many were selected
    for (int i = 0; i < m_numAttribs; i++) {
      if (group.get(i)) {
	count++;
      }
    }
    
    int[] list = new int[count];
    count = 0;
    
    for (int i = 0; i < m_numAttribs; i++) {
      if (group.get(i)) {
	list[count++] = i;
      }
    }
    
    return  list;
  }

  /**
   * generates a random subset
   * @return a random subset as a BitSet
   */
  private BitSet generateRandomSubset() {
    BitSet temp = new BitSet(m_numAttribs);
    double r;

    for (int i=0;i<m_numAttribs;i++) {
      r = m_random.nextDouble();
      if (r <= 0.5) {
	if (m_hasClass && i == m_classIndex) {
	} else {
	  temp.set(i);
	}
      }
    }
    return temp;
  }

  /**
   * counts the number of features in a subset
   * @param featureSet the feature set for which to count the features
   * @return the number of features in the subset
   */
  private int countFeatures(BitSet featureSet) {
    int count = 0;
    for (int i=0;i<m_numAttribs;i++) {
      if (featureSet.get(i)) {
	count++;
      }
    }
    return count;
  }

  /**
   * resets to defaults
   */
  private void resetOptions() {
    m_starting = null;
    m_startRange = new Range();
    m_searchSize = 0.25;
    m_seed = 1;
    m_onlyConsiderBetterAndSmaller = false;
    m_verbose = false;
  }
}