clustermembership.java

MacroWeka扩展了著名数据挖掘工具weka

/*
 *    This program is free software; you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation; either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    This program 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 General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */


/*
 *    ClusterMembership.java
 *    Copyright (C) 2004 Mark Hall
 *
 */

package weka.filters.unsupervised.attribute;

import weka.filters.Filter;
import weka.filters.UnsupervisedFilter;
import weka.filters.unsupervised.attribute.Remove;
import weka.clusterers.Clusterer;
import weka.clusterers.DensityBasedClusterer;
import weka.core.Attribute;
import weka.core.Instances;
import weka.core.Instance;
import weka.core.OptionHandler;
import weka.core.Range;
import weka.core.FastVector;
import weka.core.Option;
import weka.core.Utils;
import java.util.Enumeration;
import java.util.Vector;

/** 
 * A filter that uses a clusterer to obtain cluster membership values
 * for each input instance and outputs them as new instances. The
 * clusterer needs to be a density-based clusterer. If
 * a (nominal) class is set, then the clusterer will be run individually
 * for each class.<p>
 *
 * Valid filter-specific options are: <p>
 *
 * Full class name of clusterer to use. Clusterer options may be
 * specified at the end following a -- .(required)<p>
 *   
 * -I range string <br>
 * The range of attributes the clusterer should ignore. Note: 
 * the class attribute (if set) is automatically ignored during clustering.<p>
 *
 * @author Mark Hall ([e-mail])
 * @author Eibe Frank
 * @version $Revision: 1.1 $
 */
public class ClusterMembership extends Filter implements UnsupervisedFilter, 
							 OptionHandler {

  /** The clusterer */
  protected DensityBasedClusterer m_clusterer = new weka.clusterers.EM();

  /** Array for storing the clusterers */
  protected DensityBasedClusterer[] m_clusterers;

  /** Range of attributes to ignore */
  protected Range m_ignoreAttributesRange;

  /** Filter for removing attributes */
  protected Filter m_removeAttributes;

  /** The prior probability for each class */
  protected double[] m_priors;
  
  /**
   * 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 inputFormat can't be set successfully 
   */ 
  public boolean setInputFormat(Instances instanceInfo) throws Exception {
    
    super.setInputFormat(instanceInfo);
    m_removeAttributes = null;
    m_priors = null;

    return false;
  }

  /**
   * Signify that this batch of input to the filter is finished.
   *
   * @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 (outputFormatPeek() == null) {
      Instances toFilter = getInputFormat();
      Instances[] toFilterIgnoringAttributes;

      // Make subsets if class is nominal
      if ((toFilter.classIndex() >= 0) && toFilter.classAttribute().isNominal()) {
	toFilterIgnoringAttributes = new Instances[toFilter.numClasses()];
	for (int i = 0; i < toFilter.numClasses(); i++) {
	  toFilterIgnoringAttributes[i] = new Instances(toFilter, toFilter.numInstances());
	}
	for (int i = 0; i < toFilter.numInstances(); i++) {
	  toFilterIgnoringAttributes[(int)toFilter.instance(i).classValue()].add(toFilter.instance(i));
	}
	m_priors = new double[toFilter.numClasses()];
	for (int i = 0; i < toFilter.numClasses(); i++) {
	  toFilterIgnoringAttributes[i].compactify();
	  m_priors[i] = toFilterIgnoringAttributes[i].sumOfWeights();
	}
	Utils.normalize(m_priors);
      } else {
	toFilterIgnoringAttributes = new Instances[1];
	toFilterIgnoringAttributes[0] = toFilter;
	m_priors = new double[1];
	m_priors[0] = 1;
      }

      // filter out attributes if necessary
      if (m_ignoreAttributesRange != null || toFilter.classIndex() >= 0) {
	m_removeAttributes = new Remove();
	String rangeString = "";
	if (m_ignoreAttributesRange != null) {
	  rangeString += m_ignoreAttributesRange.getRanges();
	}
	if (toFilter.classIndex() >= 0) {
	  if (rangeString.length() > 0) {
	    rangeString += (","+(toFilter.classIndex()+1));	  
	  } else {
	    rangeString = ""+(toFilter.classIndex()+1);
	  }
	}
	((Remove)m_removeAttributes).setAttributeIndices(rangeString);
	((Remove)m_removeAttributes).setInvertSelection(false);
	((Remove)m_removeAttributes).setInputFormat(toFilter);
	for (int i = 0; i < toFilterIgnoringAttributes.length; i++) {
	  toFilterIgnoringAttributes[i] = Filter.useFilter(toFilterIgnoringAttributes[i],
							   m_removeAttributes);
	}
      }     

      // build the clusterers
      if ((toFilter.classIndex() <= 0) || !toFilter.classAttribute().isNominal()) {
	m_clusterers = DensityBasedClusterer.makeCopies(m_clusterer, 1);
	m_clusterers[0].buildClusterer(toFilterIgnoringAttributes[0]);
      } else {
	m_clusterers = DensityBasedClusterer.makeCopies(m_clusterer, toFilter.numClasses());
	for (int i = 0; i < m_clusterers.length; i++) {
	  if (toFilterIgnoringAttributes[i].numInstances() == 0) {
	    m_clusterers[i] = null;
	  } else {
	    m_clusterers[i].buildClusterer(toFilterIgnoringAttributes[i]);
	  }
	}
      }
      
      // create output dataset
      FastVector attInfo = new FastVector();
      for (int j = 0; j < m_clusterers.length; j++) {
	if (m_clusterers[j] != null) {
	  for (int i = 0; i < m_clusterers[j].numberOfClusters(); i++) {
	    attInfo.addElement(new Attribute("pCluster_" + j + "_" + i));
	  }
	}
      }
      if (toFilter.classIndex() >= 0) {
	attInfo.addElement(toFilter.classAttribute().copy());
      }
      attInfo.trimToSize();
      Instances filtered = new Instances(toFilter.relationName()+"_clusterMembership",
					 attInfo, 0);
      if (toFilter.classIndex() >= 0) {
	filtered.setClassIndex(filtered.numAttributes() - 1);
      }
      setOutputFormat(filtered);

      // build new dataset
      for (int i = 0; i < toFilter.numInstances(); i++) {
	convertInstance(toFilter.instance(i));
      }
    }
    flushInput();

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

  /**
   * 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) throws Exception {

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

    bufferInput(instance);
    return false;
  }

  /**
   * Converts logs back to density values.
   */
  protected double[] logs2densities(int j, Instance in) throws Exception {

    double[] logs = m_clusterers[j].logJointDensitiesForInstance(in);