source: sample/hadoop-0.16/tw/org/nchc/demo/DemoWordCondProb.java @ 243

/**
 * Program: DemoWordCondProb.java
 * Editor: Waue Chen 
 * From :  NCHC. Taiwn
 * Last Update Date: 07/02/2008
 * Re-code from : Cloud9: A MapReduce Library for Hadoop
 */
/*
 * Cloud9: A MapReduce Library for Hadoop
 */

package tw.org.nchc.demo;

import java.io.IOException;
import java.rmi.UnexpectedException;
import java.util.HashMap;
import java.util.Iterator;
import java.util.StringTokenizer;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.FloatWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.MapReduceBase;
import org.apache.hadoop.mapred.Mapper;
import org.apache.hadoop.mapred.OutputCollector;
import org.apache.hadoop.mapred.Partitioner;
import org.apache.hadoop.mapred.Reducer;
import org.apache.hadoop.mapred.Reporter;
import org.apache.hadoop.mapred.SequenceFileInputFormat;
import org.apache.hadoop.mapred.TextOutputFormat;
import org.apache.hadoop.mapred.lib.IdentityReducer;

import tw.org.nchc.tuple.Schema;
import tw.org.nchc.tuple.Tuple;

/**
 * <p>
 * Demo that illustrates the use of a Partitioner and special symbols in Tuple
 * to compute conditional probabilities. Demo builds on
 * {@link DemoWordCountTuple}, and has similar structure. Input comes from
 * Bible+Shakespeare sample collection, encoded as single-field tuples; see
 * {@link DemoPackRecords}. Sample of final output:
 * 
 * <pre>
 * ...
 * (admirable, *)   15.0
 * (admirable, 0)   0.6
 * (admirable, 1)   0.4
 * (admiral, *)     6.0
 * (admiral, 0)     0.33333334
 * (admiral, 1)     0.6666667
 * (admiration, *)  16.0
 * (admiration, 0)  0.625
 * (admiration, 1)  0.375
 * (admire, *)      8.0
 * (admire, 0)      0.625
 * (admire, 1)      0.375
 * (admired, *)     19.0
 * (admired, 0)     0.6315789
 * (admired, 1)     0.36842105
 * ...
 * </pre>
 * 
 * <p>
 * The first field of the key tuple contains a token. If the second field
 * contains the special symbol '*', then the value indicates the count of the
 * token in the collection. Otherwise, the value indicates p(EvenOrOdd|Token),
 * the probability that a line is odd-length or even-length, given the
 * occurrence of a token.
 * </p>
 */
public class DemoWordCondProb {

  // create the schema for the tuple that will serve as the key
  private static final Schema KEY_SCHEMA = new Schema();

  // define the schema statically
  static {
    KEY_SCHEMA.addField("Token", String.class, "");
    KEY_SCHEMA.addField("EvenOrOdd", Integer.class, new Integer(1));
  }

  // mapper that emits tuple as the key, and value '1' for each occurrence
  private static class MapClass extends MapReduceBase implements
      Mapper<LongWritable, Tuple, Tuple, FloatWritable> {
    private final static FloatWritable one = new FloatWritable(1);
    private Tuple tupleOut = KEY_SCHEMA.instantiate();

    public void map(LongWritable key, Tuple tupleIn,
        OutputCollector<Tuple, FloatWritable> output, Reporter reporter)
        throws IOException {

      // the input value is a tuple; get field 0
      // see DemoPackRecords of how input SequenceFile is generated
      String line = (String) ((Tuple) tupleIn).get(0);
      StringTokenizer itr = new StringTokenizer(line);
      while (itr.hasMoreTokens()) {
        String token = itr.nextToken();

        // emit key-value pair for either even-length or odd-length line
        tupleOut.set("Token", token);
        tupleOut.set("EvenOrOdd", line.length() % 2);
        output.collect(tupleOut, one);

        // emit key-value pair for the total count
        tupleOut.set("Token", token);
        // use special symbol in field 2
        tupleOut.setSymbol("EvenOrOdd", "*");
        output.collect(tupleOut, one);
      }
    }
  }

  // reducer computes conditional probabilities
  private static class ReduceClass extends MapReduceBase implements
      Reducer<Tuple, FloatWritable, Tuple, FloatWritable> {
    // HashMap keeps track of total counts
    private final static HashMap<String, Integer> TotalCounts = new HashMap<String, Integer>();

    public synchronized void reduce(Tuple tupleKey,
        Iterator<FloatWritable> values,
        OutputCollector<Tuple, FloatWritable> output, Reporter reporter)
        throws IOException {
      // sum values
      int sum = 0;
      while (values.hasNext()) {
        sum += values.next().get();
      }

      String tok = (String) tupleKey.get("Token");

      // check if the second field is a special symbol
      if (tupleKey.containsSymbol("EvenOrOdd")) {
        // emit total count
        output.collect(tupleKey, new FloatWritable(sum));
        // record total count
        TotalCounts.put(tok, sum);
      } else {
        if (!TotalCounts.containsKey(tok))
          throw new UnexpectedException("Don't have total counts!");

        // divide sum by total count to obtain conditional probability
        float p = (float) sum / TotalCounts.get(tok);

        // emit P(EvenOrOdd|Token)
        output.collect(tupleKey, new FloatWritable(p));
      }
    }
  }

  // partition by first field of the tuple, so that tuples corresponding
  // to the same token will be sent to the same reducer
  private static class MyPartitioner implements
      Partitioner<Tuple, FloatWritable> {
    public void configure(JobConf job) {
    }

    public int getPartition(Tuple key, FloatWritable value,
        int numReduceTasks) {
      return (key.get("Token").hashCode() & Integer.MAX_VALUE)
          % numReduceTasks;
    }
  }

  // dummy constructor
  private DemoWordCondProb() {
  }

  /**
   * Runs the demo.
   */
  public static void main(String[] args) throws IOException {
    String inPath = "/shared/sample-input/bible+shakes.nopunc.packed";
    String output1Path = "condprob";
    int numMapTasks = 20;
    int numReduceTasks = 10;

    // first MapReduce cycle is to do the tuple counting
    JobConf conf1 = new JobConf(DemoWordCondProb.class);
    conf1.setJobName("DemoWordCondProb.MR1");

    conf1.setNumMapTasks(numMapTasks);
    conf1.setNumReduceTasks(numReduceTasks);
    conf1.setInputPath(new Path(inPath));
    
    conf1.setInputFormat(SequenceFileInputFormat.class);
    
    conf1.setOutputPath(new Path(output1Path));
    conf1.setOutputKeyClass(Tuple.class);
    conf1.setOutputValueClass(FloatWritable.class);
    conf1.setOutputFormat(TextOutputFormat.class);

    conf1.setMapperClass(MapClass.class);
    // this is a potential gotcha! can't use ReduceClass for combine because
    // we have not collected all the counts yet, so we can't divide through
    // to compute the conditional probabilities
    conf1.setCombinerClass(IdentityReducer.class);
    conf1.setReducerClass(ReduceClass.class);
    conf1.setPartitionerClass(MyPartitioner.class);

    JobClient.runJob(conf1);
  }
}