subsetsizeforwardselection.java

这是关于数据挖掘的一些算法

    return "Set the type of the search.";
  }

  /**
   * Set the type
   *
   * @param t
   *            the Linear Forward Selection type
   */
  public void setType(SelectedTag t) {
    if (t.getTags() == TAGS_TYPE) {
      m_linearSelectionType = t.getSelectedTag().getID();
    }
  }

  /**
   * Get the type
   *
   * @return the Linear Forward Selection type
   */
  public SelectedTag getType() {
    return new SelectedTag(m_linearSelectionType, TAGS_TYPE);
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for displaying in the
   *         explorer/experimenter gui
   */
  public String subsetSizeEvaluatorTipText() {
    return "Subset evaluator to use for subset size determination.";
  }

  /**
   * Set the subset evaluator to use for subset size determination.
   *
   * @param eval
   *            the subset evaluator to use for subset size determination.
   */
  public void setSubsetSizeEvaluator(ASEvaluation eval)
    throws Exception {
    if (!SubsetEvaluator.class.isInstance(eval)) {
      throw new Exception(eval.getClass().getName() +
                          " is no subset evaluator.");
    }

    m_setSizeEval = (SubsetEvaluator) eval;
  }

  /**
   * Get the subset evaluator used for subset size determination.
   *
   * @return the evaluator used for subset size determination.
   */
  public ASEvaluation getSubsetSizeEvaluator() {
    return m_setSizeEval;
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for displaying in the
   *         explorer/experimenter gui
   */
  public String numSubsetSizeCVFoldsTipText() {
    return "Number of cross validation folds for subset size determination";
  }

  /**
   * Set the number of cross validation folds for subset size determination
   * (default = 5).
   *
   * @param f
   *            number of folds
   */
  public void setNumSubsetSizeCVFolds(int f) {
    m_numFolds = f;
  }

  /**
   * Get the number of cross validation folds for subset size determination
   * (default = 5).
   *
   * @return number of folds
   */
  public int getNumSubsetSizeCVFolds() {
    return m_numFolds;
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for displaying in the
   *         explorer/experimenter gui
   */
  public String seedTipText() {
    return "Seed for cross validation subset size determination. (default = 1)";
  }

  /**
   * Seed for cross validation subset size determination. (default = 1)
   *
   * @param s
   *            seed
   */
  public void setSeed(int s) {
    m_seed = s;
  }

  /**
   * Seed for cross validation subset size determination. (default = 1)
   *
   * @return seed
   */
  public int getSeed() {
    return m_seed;
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for displaying in the
   *         explorer/experimenter gui
   */
  public String verboseTipText() {
    return "Turn on verbose output for monitoring the search's progress.";
  }

  /**
   * Set whether verbose output should be generated.
   *
   * @param d
   *            true if output is to be verbose.
   */
  public void setVerbose(boolean b) {
    m_verbose = b;
  }

  /**
   * Get whether output is to be verbose
   *
   * @return true if output will be verbose
   */
  public boolean getVerbose() {
    return m_verbose;
  }

  /**
   * Gets the current settings of LinearForwardSelection.
   *
   * @return an array of strings suitable for passing to setOptions()
   */
  public String[] getOptions() {
    String[] evaluatorOptions = new String[0];

    if ((m_setSizeEval != null) && (m_setSizeEval instanceof OptionHandler)) {
      evaluatorOptions = ((OptionHandler) m_setSizeEval).getOptions();
    }

    String[] options = new String[15 + evaluatorOptions.length];
    int current = 0;

    if (m_performRanking) {
      options[current++] = "-I";
    }

    options[current++] = "-K";
    options[current++] = "" + m_numUsedAttributes;
    options[current++] = "-T";
    options[current++] = "" + m_linearSelectionType;

    options[current++] = "-F";
    options[current++] = "" + m_numFolds;
    options[current++] = "-S";
    options[current++] = "" + m_seed;

    options[current++] = "-Z";
    options[current++] = "" + m_verbose;

    if (m_setSizeEval != null) {
      options[current++] = "-E";
      options[current++] = m_setSizeEval.getClass().getName();
    }

    options[current++] = "--";
    System.arraycopy(evaluatorOptions, 0, options, current,
                     evaluatorOptions.length);
    current += evaluatorOptions.length;

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

    return options;
  }

  /**
   * returns a description of the search as a String
   *
   * @return a description of the search
   */
  public String toString() {
    StringBuffer LFSString = new StringBuffer();

    LFSString.append("\tSubset Size Forward Selection.\n");

    LFSString.append("\tLinear Forward Selection Type: ");

    if (m_linearSelectionType == TYPE_FIXED_SET) {
      LFSString.append("fixed-set\n");
    } else {
      LFSString.append("fixed-width\n");
    }

    LFSString.append("\tNumber of top-ranked attributes that are used: " +
                     m_numUsedAttributes + "\n");

    LFSString.append(
                     "\tNumber of cross validation folds for subset size determination: " +
                     m_numFolds + "\n");
    LFSString.append("\tSeed for cross validation subset size determination: " +
                     m_seed + "\n");

    LFSString.append("\tTotal number of subsets evaluated: " + m_totalEvals +
                     "\n");
    LFSString.append("\tMerit of best subset found: " +
                     Utils.doubleToString(Math.abs(m_bestMerit), 8, 3) + "\n");

    return LFSString.toString();
  }

  /**
   * Searches the attribute subset space by subset size forward selection
   *
   * @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 {
    m_totalEvals = 0;

    if (!(ASEval instanceof SubsetEvaluator)) {
      throw new Exception(ASEval.getClass().getName() + " is not a " +
                          "Subset evaluator!");
    }

    if (m_setSizeEval == null) {
      m_setSizeEval = (SubsetEvaluator) ASEval;
    }

    m_numAttribs = data.numAttributes();

    if (m_numUsedAttributes > m_numAttribs) {
      System.out.println(
                         "Decreasing number of top-ranked attributes to total number of attributes: " +
                         data.numAttributes());
      m_numUsedAttributes = m_numAttribs;
    }

    Instances[] trainData = new Instances[m_numFolds];
    Instances[] testData = new Instances[m_numFolds];
    LFSMethods[] searchResults = new LFSMethods[m_numFolds];

    Random random = new Random(m_seed);
    Instances dataCopy = new Instances(data);
    dataCopy.randomize(random);

    if (dataCopy.classAttribute().isNominal()) {
      dataCopy.stratify(m_numFolds);
    }

    for (int f = 0; f < m_numFolds; f++) {
      trainData[f] = dataCopy.trainCV(m_numFolds, f, random);
      testData[f] = dataCopy.testCV(m_numFolds, f);
    }

    LFSMethods LSF = new LFSMethods();

    int[] ranking;

    if (m_performRanking) {
      ((SubsetEvaluator) ASEval).buildEvaluator(data);
      ranking = LSF.rankAttributes(data, (SubsetEvaluator) ASEval, m_verbose);
    } else {
      ranking = new int[m_numAttribs];

      for (int i = 0; i < ranking.length; i++) {
        ranking[i] = i;
      }
    }

    int maxSubsetSize = 0;

    for (int f = 0; f < m_numFolds; f++) {
      if (m_verbose) {
        System.out.println("perform search on internal fold: " + (f + 1) + "/" +
                           m_numFolds);
      }

      m_setSizeEval.buildEvaluator(trainData[f]);
      searchResults[f] = new LFSMethods();
      searchResults[f].forwardSearch(m_cacheSize, new BitSet(m_numAttribs),
                                     ranking, m_numUsedAttributes,
                                     m_linearSelectionType == TYPE_FIXED_WIDTH, 1, -1, trainData[f],
                                     m_setSizeEval, m_verbose);

      maxSubsetSize = Math.max(maxSubsetSize,
                               searchResults[f].getBestGroup().cardinality());
    }

    if (m_verbose) {
      System.out.println(
                         "continue searches on internal folds to maxSubsetSize (" +
                         maxSubsetSize + ")");
    }

    for (int f = 0; f < m_numFolds; f++) {
      if (m_verbose) {
        System.out.print("perform search on internal fold: " + (f + 1) + "/" +
                         m_numFolds + " with starting set ");
        LFSMethods.printGroup(searchResults[f].getBestGroup(),
                              trainData[f].numAttributes());
      }

      if (searchResults[f].getBestGroup().cardinality() < maxSubsetSize) {
        m_setSizeEval.buildEvaluator(trainData[f]);
        searchResults[f].forwardSearch(m_cacheSize,
                                       searchResults[f].getBestGroup(), ranking, m_numUsedAttributes,
                                       m_linearSelectionType == TYPE_FIXED_WIDTH, 1, maxSubsetSize,
                                       trainData[f], m_setSizeEval, m_verbose);
      }
    }

    double[][] testMerit = new double[m_numFolds][maxSubsetSize + 1];

    for (int f = 0; f < m_numFolds; f++) {
      for (int s = 1; s <= maxSubsetSize; s++) {
        if (HoldOutSubsetEvaluator.class.isInstance(m_setSizeEval)) {
          m_setSizeEval.buildEvaluator(trainData[f]);
          testMerit[f][s] = ((HoldOutSubsetEvaluator) m_setSizeEval).evaluateSubset(searchResults[f].getBestGroupOfSize(
                                                                                                                        s), testData[f]);
        } else {
          m_setSizeEval.buildEvaluator(testData[f]);
          testMerit[f][s] = m_setSizeEval.evaluateSubset(searchResults[f].getBestGroupOfSize(
                                                                                             s));
        }
      }
    }

    double[] avgTestMerit = new double[maxSubsetSize + 1];
    int finalSubsetSize = -1;

    for (int s = 1; s <= maxSubsetSize; s++) {
      for (int f = 0; f < m_numFolds; f++) {
        avgTestMerit[s] = ((avgTestMerit[s] * f) + testMerit[f][s]) / (double) (f +
                                                                                1);
      }

      if ((finalSubsetSize == -1) ||
          (avgTestMerit[s] > avgTestMerit[finalSubsetSize])) {
        finalSubsetSize = s;
      }

      if (m_verbose) {
        System.out.println("average merit for subset-size " + s + ": " +
                           avgTestMerit[s]);
      }
    }

    if (m_verbose) {
      System.out.println("performing final forward selection to subset-size: " +
                         finalSubsetSize);
    }

    ((SubsetEvaluator) ASEval).buildEvaluator(data);
    LSF.forwardSearch(m_cacheSize, new BitSet(m_numAttribs), ranking,
                      m_numUsedAttributes, m_linearSelectionType == TYPE_FIXED_WIDTH, 1,
                      finalSubsetSize, data, (SubsetEvaluator) ASEval, m_verbose);

    m_totalEvals = LSF.getNumEvalsTotal();
    m_bestMerit = LSF.getBestMerit();

    return attributeList(LSF.getBestGroup());
  }

  /**
   * Reset options to default values
   */
  protected void resetOptions() {
    m_performRanking = true;
    m_numUsedAttributes = 50;
    m_linearSelectionType = TYPE_FIXED_SET;
    m_setSizeEval = new ClassifierSubsetEval();
    m_numFolds = 5;
    m_seed = 1;
    m_totalEvals = 0;
    m_cacheSize = 1;
    m_verbose = false;
  }

  /**
   * converts a BitSet into a list of attribute indexes
   *
   * @param group
   *            the BitSet to convert
   * @return an array of attribute indexes
   */
  protected 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;
  }
}