twopassdataindexer.java

最大熵分类器

///////////////////////////////////////////////////////////////////////////////
//Copyright (C) 2003 Thomas Morton
//
//This library is free 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 2.1 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 program; if not, write to the Free Software
//Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
//////////////////////////////////////////////////////////////////////////////   
package opennlp.maxent;

import gnu.trove.TIntArrayList;
import gnu.trove.TObjectIntHashMap;

import java.io.BufferedWriter;
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.OutputStreamWriter;
import java.io.Writer;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
import java.util.Set;

/**
 * Collecting event and context counts by making two passes over the events.  The
 * first pass determines which contexts will be used by the model, and the
 * second pass creates the events in memory containing only the contexts which 
 * will be used.  This greatly reduces the amount of memory required for storing
 * the events.  During the first pass a temporary event file is created which
 * is read during the second pass.
 */
public class TwoPassDataIndexer extends AbstractDataIndexer{

  /**
   * One argument constructor for DataIndexer which calls the two argument
   * constructor assuming no cutoff.
   *
   * @param eventStream An Event[] which contains the a list of all the Events
   *               seen in the training data.
   */
  public TwoPassDataIndexer(EventStream eventStream) throws IOException {
    this(eventStream, 0);
  }

  /**
   * Two argument constructor for DataIndexer.
   *
   * @param eventStream An Event[] which contains the a list of all the Events
   *               seen in the training data.
   * @param cutoff The minimum number of times a predicate must have been
   *               observed in order to be included in the model.
   */
  public TwoPassDataIndexer(EventStream eventStream, int cutoff) throws IOException {
    TObjectIntHashMap predicateIndex;
    List eventsToCompare;

    predicateIndex = new TObjectIntHashMap();
    System.out.println("Indexing events using cutoff of " + cutoff + "\n");

    System.out.print("\tComputing event counts...  ");
    try {
      File tmp = File.createTempFile("events", null);
      tmp.deleteOnExit();
      Writer osw = new BufferedWriter(new OutputStreamWriter(new FileOutputStream(tmp),"UTF8"));
      int numEvents = computeEventCounts(eventStream, osw, predicateIndex, cutoff);
      System.out.println("done. " + numEvents + " events");

      System.out.print("\tIndexing...  ");

      eventsToCompare = index(numEvents, new FileEventStream(tmp), predicateIndex);
      // done with predicates
      predicateIndex = null;
      tmp.delete();
      System.out.println("done.");

      System.out.print("Sorting and merging events... ");
      sortAndMerge(eventsToCompare);
      System.out.println("Done indexing.");
    }
    catch(IOException e) {
      System.err.println(e);
    }
  }

  /**
      * Reads events from <tt>eventStream</tt> into a linked list.  The
      * predicates associated with each event are counted and any which
      * occur at least <tt>cutoff</tt> times are added to the
      * <tt>predicatesInOut</tt> map along with a unique integer index.
      *
      * @param eventStream an <code>EventStream</code> value
      * @param eventStore a writer to which the events are written to for later processing.
      * @param predicatesInOut a <code>TObjectIntHashMap</code> value
      * @param cutoff an <code>int</code> value
      */
  private int computeEventCounts(EventStream eventStream, Writer eventStore, TObjectIntHashMap predicatesInOut, int cutoff) throws IOException {
    TObjectIntHashMap counter = new TObjectIntHashMap();
    int eventCount = 0;
    Set predicateSet = new HashSet();
    while (eventStream.hasNext()) {
      Event ev = eventStream.nextEvent();
      eventCount++;
      eventStore.write(FileEventStream.toLine(ev));
      String[] ec = ev.getContext();
      update(ec,predicateSet,counter,cutoff);
    }
    predCounts = new int[predicateSet.size()];
    int index = 0;
    for (Iterator pi=predicateSet.iterator();pi.hasNext();index++) {
      String predicate = (String) pi.next();
      predCounts[index] = counter.get(predicate);
      predicatesInOut.put(predicate,index);
    }
    eventStore.close();
    return eventCount;
  }

  private List index(int numEvents, EventStream es, TObjectIntHashMap predicateIndex) {
    TObjectIntHashMap omap = new TObjectIntHashMap();
    int outcomeCount = 0;
    List eventsToCompare = new ArrayList(numEvents);
    TIntArrayList indexedContext = new TIntArrayList();
    while (es.hasNext()) {
      Event ev = es.nextEvent();
      String[] econtext = ev.getContext();
      ComparableEvent ce;

      int ocID;
      String oc = ev.getOutcome();

      if (omap.containsKey(oc)) {
        ocID = omap.get(oc);
      }
      else {
        ocID = outcomeCount++;
        omap.put(oc, ocID);
      }

      for (int i = 0; i < econtext.length; i++) {
        String pred = econtext[i];
        if (predicateIndex.containsKey(pred)) {
          indexedContext.add(predicateIndex.get(pred));
        }
      }

      // drop events with no active features
      if (indexedContext.size() > 0) {
        ce = new ComparableEvent(ocID, indexedContext.toNativeArray());
        eventsToCompare.add(ce);
      }
      else {
        System.err.println("Dropped event " + ev.getOutcome() + ":" + Arrays.asList(ev.getContext()));
      }
      // recycle the TIntArrayList
      indexedContext.resetQuick();
    }
    outcomeLabels = toIndexedStringArray(omap);
    predLabels = toIndexedStringArray(predicateIndex);
    return eventsToCompare;
  }

}