discretizefilter.java

一个数据挖掘系统的源码

/**
 *
 *   AgentAcademy - an open source Data Mining framework for
 *   training intelligent agents
 *
 *   Copyright (C)   2001-2003 AA Consortium.
 *
 *   This library is open source software; you can redistribute it
 *   and/or modify it under the terms of the GNU Lesser General
 *   Public License as published by the Free Software Foundation;
 *   either version 2.0 of the License, or (at your option) any later
 *   version.
 *
 *   This library is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *   GNU General Public License for more details.
 *
 *   You should have received a copy of the GNU Lesser General Public
 *   License along with this library; if not, write to the Free
 *   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston,
 *   MA  02111-1307 USA
 *
 */

package org.agentacademy.modules.dataminer.filters;

import java.util.Enumeration;
import java.util.Vector;

import org.agentacademy.modules.dataminer.core.Attribute;
import org.agentacademy.modules.dataminer.core.ContingencyTables;
import org.agentacademy.modules.dataminer.core.FastVector;
import org.agentacademy.modules.dataminer.core.Instance;
import org.agentacademy.modules.dataminer.core.Instances;
import org.agentacademy.modules.dataminer.core.Option;
import org.agentacademy.modules.dataminer.core.OptionHandler;
import org.agentacademy.modules.dataminer.core.Range;
import org.agentacademy.modules.dataminer.core.SparseInstance;
import org.agentacademy.modules.dataminer.core.SpecialFunctions;
import org.agentacademy.modules.dataminer.core.UnassignedClassException;
import org.agentacademy.modules.dataminer.core.UnsupportedClassTypeException;
import org.agentacademy.modules.dataminer.core.Utils;
import org.agentacademy.modules.dataminer.core.WeightedInstancesHandler;
import org.apache.log4j.Logger;

/**
 * An instance filter that discretizes a range of numeric attributes in
 * the dataset into nominal attributes. Discretization can be either by
 * simple binning, or by Fayyad & Irani's MDL method (the default).<p>
 *
 * Valid filter-specific options are: <p>
 *
 * -B num <br>
 * Specifies the (maximum) number of bins to divide numeric attributes into.
 * (default: class-based discretisation).<p>
 *
 * -F <br>
 * Use equal-frequency instead of equal-width discretization if
 * class-based discretisation is turned off.<p>
 *
 * -O <br>
 * Optimize the number of bins using a leave-one-out estimate of the
 * entropy (for equal-width binning).<p>
 *
 * -R col1,col2-col4,... <br>
 * Specifies list of columns to Discretize. First
 * and last are valid indexes. (default: none) <p>
 *
 * -V <br>
 * Invert matching sense.<p>
 *
 * -D <br>
 * Make binary nominal attributes. <p>
 *
 * -E <br>
 * Use better encoding of split point for MDL. <p>
 *
 * -K <br>
 * Use Kononeko's MDL criterion. <p>
 *
 * @author Len Trigg ([e-mail])
 * @author Eibe Frank ([e-mail])
 * @version $Revision: 1.3 $
 */
public class DiscretizeFilter extends Filter
  implements OptionHandler, WeightedInstancesHandler {

  public static Logger                log = Logger.getLogger(DiscretizeFilter.class);
  /** Stores which columns to Discretize */
  protected Range m_DiscretizeCols = new Range();

  /** The number of bins to divide the attribute into */
  protected int m_NumBins = 10;

  /** Store the current cutpoints */
  protected double [][] m_CutPoints = null;

  /** True if discretisation will be done by MDL rather than binning */
  protected boolean m_UseMDL = true;

  /** Output binary attributes for discretized attributes. */
  protected boolean m_MakeBinary = false;

  /** Use better encoding of split point for MDL. */
  protected boolean m_UseBetterEncoding = false;

  /** Use Kononenko's MDL criterion instead of Fayyad et al.'s */
  protected boolean m_UseKononenko = false;

  /** Find the number of bins using cross-validated entropy. */
  protected boolean m_FindNumBins = false;

  /** Use equal-frequency binning if unsupervised discretization turned on */
  protected boolean m_UseEqualFrequency = false;

  /** Constructor - initialises the filter */
  public DiscretizeFilter() {

    setAttributeIndices("first-last");
  }


  /**
   * Gets an enumeration describing the available options.
   *
   * @return an enumeration of all the available options.
   */
  public Enumeration listOptions() {

    Vector newVector = new Vector(7);

    newVector.addElement(new Option(
              "\tSpecifies the (maximum) number of bins to divide numeric"
	      + " attributes into.\n"
	      + "\t(default class-based discretization)",
              "B", 1, "-B <num>"));

    newVector.addElement(new Option(
              "\tUse equal-frequency instead of equal-width with\n"+
	      "\tunsupervised discretization.",
              "F", 0, "-F"));

    newVector.addElement(new Option(
              "\tOptimize number of bins using leave-one-out estimate\n"+
	      "\tof estimated entropy (for equal-width discretization).",
              "O", 0, "-O"));

    /* If we decide to implement loading and saving cutfiles like
     * the C Discretizer (which is probably not necessary)
    newVector.addElement(new Option(
              "\tSpecify that the cutpoints should be loaded from a file.",
              "L", 1, "-L <file>"));
    newVector.addElement(new Option(
              "\tSpecify that the chosen cutpoints should be saved to a file.",
              "S", 1, "-S <file>"));
    */

    newVector.addElement(new Option(
              "\tSpecifies list of columns to Discretize. First"
	      + " and last are valid indexes.\n"
	      + "\t(default none)",
              "R", 1, "-R <col1,col2-col4,...>"));

    newVector.addElement(new Option(
              "\tInvert matching sense of column indexes.",
              "V", 0, "-V"));

    newVector.addElement(new Option(
              "\tOutput binary attributes for discretized attributes.",
              "D", 0, "-D"));

    newVector.addElement(new Option(
              "\tUse better encoding of split point for MDL.",
              "E", 0, "-E"));

    newVector.addElement(new Option(
              "\tUse Kononenko's MDL criterion.",
              "K", 0, "-K"));

    return newVector.elements();
  }


  /**
   * Parses the options for this object. Valid options are: <p>
   *
   * -B num <br>
   * Specifies the (maximum) number of bins to divide numeric attributes into.
   * (default class-based discretisation).<p>
   *
   * -F <br>
   * Use equal-frequency instead of equal-width discretization if
   * class-based discretisation is turned off.<p>
   *
   * -O <br>
   * Optimize the number of bins using a leave-one-out estimate of the
   * entropy (for equal-width binning).<p>
   *
   * -R col1,col2-col4,... <br>
   * Specifies list of columns to Discretize. First
   * and last are valid indexes. (default none) <p>
   *
   * -V <br>
   * Invert matching sense.<p>
   *
   * -D <br>
   * Make binary nominal attributes. <p>
   *
   * -E <br>
   * Use better encoding of split point for MDL. <p>
   *
   * -K <br>
   * Use Kononeko's MDL criterion. <p>
   *
   * @param options the list of options as an array of strings
   * @exception Exception if an option is not supported
   */
  public void setOptions(String[] options) throws Exception {

    setMakeBinary(Utils.getFlag('D', options));
    setUseEqualFrequency(Utils.getFlag('F', options));
    setUseBetterEncoding(Utils.getFlag('E', options));
    setUseKononenko(Utils.getFlag('K', options));
    setFindNumBins(Utils.getFlag('O', options));
    setInvertSelection(Utils.getFlag('V', options));
    setUseMDL(true);

    String numBins = Utils.getOption('B', options);
    if (numBins.length() != 0) {
      setBins(Integer.parseInt(numBins));
      setUseMDL(false);
    } else {
      setBins(10);
    }

    String convertList = Utils.getOption('R', options);
    if (convertList.length() != 0) {
      setAttributeIndices(convertList);
    } else {
      setAttributeIndices("first-last");
    }

    if (getInputFormat() != null) {
      setInputFormat(getInputFormat());
    }
  }
  /**
   * Gets the current settings of the filter.
   *
   * @return an array of strings suitable for passing to setOptions
   */
  public String [] getOptions() {

    String [] options = new String [12];
    int current = 0;

    if (getMakeBinary()) {
      options[current++] = "-D";
    }
    if (getUseEqualFrequency()) {
      options[current++] = "-F";
    }
    if (getUseBetterEncoding()) {
      options[current++] = "-E";
    }
    if (getUseKononenko()) {
      options[current++] = "-K";
    }
    if (getFindNumBins()) {
      options[current++] = "-O";
    }
    if (getInvertSelection()) {
      options[current++] = "-V";
    }
    if (!getUseMDL()) {
      options[current++] = "-B"; options[current++] = "" + getBins();
    }
    if (!getAttributeIndices().equals("")) {
      options[current++] = "-R"; options[current++] = getAttributeIndices();
    }
    while (current < options.length) {
      options[current++] = "";
    }
    return options;
  }


  /**
   * Sets the format of the input instances.
   *
   * @param instanceInfo an Instances object containing the input instance
   * structure (any instances contained in the object are ignored - only the
   * structure is required).
   * @return true if the outputFormat may be collected immediately
   * @exception Exception if the input format can't be set successfully
   */
  public boolean setInputFormat(Instances instanceInfo) throws Exception {

    super.setInputFormat(instanceInfo);

    m_DiscretizeCols.setUpper(instanceInfo.numAttributes() - 1);
    m_CutPoints = null;
    if (m_UseMDL) {
      if (instanceInfo.classIndex() < 0) {
	throw new UnassignedClassException("Cannot use class-based discretization: "
                                           + "no class assigned to the dataset");
      }
      if (!instanceInfo.classAttribute().isNominal()) {
	throw new UnsupportedClassTypeException("Supervised discretization not possible:"
                                                + " class is not nominal!");
      }
    } else {
      if (getFindNumBins() && getUseEqualFrequency()) {
	throw new IllegalArgumentException("Bin number optimization in conjunction "+
					   "with equal-frequency binning not implemented.");
      }
    }

    // If we implement loading cutfiles, then load
    //them here and set the output format
    return false;
  }



  /**
   * Input an instance for filtering. Ordinarily the instance is processed
   * and made available for output immediately. Some filters require all
   * instances be read before producing output.
   *
   * @param instance the input instance
   * @return true if the filtered instance may now be
   * collected with output().
   * @exception IllegalStateException if no input format has been defined.
   */
  public boolean input(Instance instance) {

    if (getInputFormat() == null) {
      throw new IllegalStateException("No input instance format defined");
    }
    if (m_NewBatch) {
      resetQueue();
      m_NewBatch = false;
    }

    if (m_CutPoints != null) {
      convertInstance(instance);
      return true;
    }

    bufferInput(instance);
    return false;
  }


  /**
   * Signifies that this batch of input to the filter is finished. If the
   * filter requires all instances prior to filtering, output() may now
   * be called to retrieve the filtered instances.
   *
   * @return true if there are instances pending output
   * @exception IllegalStateException if no input structure has been defined
   */
  public boolean batchFinished() throws Exception {

    if (getInputFormat() == null) {
      throw new IllegalStateException("No input instance format defined");
    }
    if (m_CutPoints == null) {
      calculateCutPoints();

      setOutputFormat();

      // If we implement saving cutfiles, save the cuts here

      // Convert pending input instances
      for(int i = 0; i < getInputFormat().numInstances(); i++) {
	convertInstance(getInputFormat().instance(i));
      }
    }
    flushInput();

    m_NewBatch = true;
    return (numPendingOutput() != 0);
  }

  /**
   * Returns a string describing this filter
   *
   * @return a description of the filter suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {

    return "An instance filter that discretizes a range of numeric"
      + " attributes in the dataset into nominal attributes."
      + " Discretization can be either by simple binning, or by"
      + " Fayyad & Irani's MDL method (the default).";
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String findNumBinsTipText() {

    return "Optimize number of equal-width bins using leave-one-out.";
  }

  /**
   * Get the value of FindNumBins.
   *
   * @return Value of FindNumBins.
   */
  public boolean getFindNumBins() {

    return m_FindNumBins;
  }

  /**
   * Set the value of FindNumBins.
   *
   * @param newFindNumBins Value to assign to FindNumBins.
   */
  public void setFindNumBins(boolean newFindNumBins) {

    m_FindNumBins = newFindNumBins;
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for