onepassdataindexer.java~8~

垃圾邮件过滤器源代码

///////////////////////////////////////////////////////////////////////////////
// Copyright (C) 2001 Jason Baldridge and Gann Bierner
//
// 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.*;
import java.util.*;

/**
 * An indexer for maxent model data which handles cutoffs for uncommon
 * contextual predicates and provides a unique integer index for each of the
 * predicates.  The data structures built in the constructor of this class are
 * used by the GIS trainer.
 *
 * @author      Jason Baldridge
 * @version $Revision: 1.1 $, $Date: 2003/12/13 16:41:29 $
 */
public class OnePassDataIndexer 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 OnePassDataIndexer(EventStream eventStream) {
        this(eventStream, 0);
    }

    /**
     * Two argument constructor for DataIndexer.
     * eventStream为训练数据的事件集,cutoff为特征起作用的最小次数
     * @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.
     */
   //自定义修改的EventStream可能要在此修改?
   public OnePassDataIndexer(EventStream eventStream, int cutoff) {
     TObjectIntHashMap predicateIndex;
     TDoubleIntHashMap predicateWeightIndex;
     TLinkedList events;
     List eventsToCompare;

     predicateIndex = new TObjectIntHashMap();
     predicateWeightIndex = new TDoubleIntHashMap();

     System.out.println("\nIndexing events using cutoff of " + cutoff + "\n");

     System.out.print("\tComputing event counts...  ");
     events = computeEventCounts(eventStream, predicateIndex,
                                 predicateWeightIndex, cutoff);
     System.out.println("done. " + events.size() + " events");
     // 对events的计算是正确的,下面作修改

     System.out.print("\tIndexing...  ");
     eventsToCompare = index(events, predicateIndex, predicateWeightIndex);
     // done with event list 以上部分正确
     events = null;
     // done with predicates
     predicateIndex = null;
     predicateWeightIndex = null;
     System.out.println("done.");
     System.out.print("Sorting and merging events... ");
     sortAndMerge(eventsToCompare);
     System.out.println("Done indexing.");

   }



    /**
     * 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.
     * 从事件流中把事件读入链表,计算每个事件的断言个数(>=cutoff),加入到HASH表中
     * @param eventStream an <code>EventStream</code> value
     * @param predicatesInOut a <code>TObjectIntHashMap</code> value
     * @param cutoff an <code>int</code> value
     * @return a <code>TLinkedList</code> value
     */
    private TLinkedList computeEventCounts(EventStream eventStream,
                                           TObjectIntHashMap predicatesInOut,
                                           TDoubleIntHashMap
                                           predicatesWeightInOut,
                                           int cutoff) {

      TObjectIntHashMap counter = new TObjectIntHashMap();
      TLinkedList events = new TLinkedList();
      int predicateIndex = 0;
      int predicateWeightIndex = 0;

      while (eventStream.hasNext()) {
        Event ev = eventStream.nextEvent(); //取得下一个事件
        events.addLast(ev); //加到事件表中
        // 分析当前事件的各个断言是否已经出现在索引表中
        Predicate[] preds = ev.getMailContext();
        for (int j = 0; j < preds.length; j++) {
          // 处理断言字符串的索引表
          if (!predicatesInOut.containsKey(preds[j].word)) { //没有出现,应该加入
            if (counter.increment(preds[j].word)) { // 次数++,只有次数大于cutoff的才加入索引表中使用
            }
            else {
              counter.put(preds[j].word, 1); //置初始值为1
            }
            if (counter.get(preds[j].word) >= cutoff) { // 次数大于cutoff,加入到索引表中
              predicatesInOut.put(preds[j].word, predicateIndex++);
              counter.remove(preds[j].word);
            }
          }

          //处理权重的索引表,权重不需要考虑次数,因此不需要设 counter
          if (!predicatesWeightInOut.containsKey(preds[j].weight)) // 此权重没有出现在索引表里,直接加入
            predicatesWeightInOut.put(preds[j].weight, predicateWeightIndex++);
        }
      }

    predicatesInOut.trimToSize();
    predicatesWeightInOut.trimToSize();

    return events;
  }

  private List index(TLinkedList events,
                     TObjectIntHashMap predicateIndex,
                     TDoubleIntHashMap predicateWeightIndex) {

    TObjectIntHashMap omap = new TObjectIntHashMap();
    TIntArrayList indexedContext = new TIntArrayList();
    TIntArrayList indexedWeight = new TIntArrayList();

    List eventsToCompare = new ArrayList(events.size());

    int numEvents = events.size();
    int outcomeCount = 0;
    int predCount = 0;

    for (int eventIndex = 0; eventIndex < numEvents; eventIndex++) {
      Event ev = (Event) events.removeFirst();
      Predicate[] Preds = ev.getMailContext();
      ComparableEvent ce;

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

      // 处理输出结果
      if (omap.containsKey(oc)) {
        ocID = omap.get(oc);
      }
      else {
        ocID = outcomeCount++;
        omap.put(oc, ocID);
      }
      // 处理正文
      for (int i = 0; i < Preds.length; i++) {
        String pred = Preds[i].word;
        double weight = Preds[i].weight;
        // if (predicateIndex.containsKey(pred)) {
        indexedContext.add(predicateIndex.get(pred));
        indexedWeight.add(predicateWeightIndex.get(weight));
        // }
      }

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