矩阵的乘法只有在第一个矩阵的列数(column)和第二个矩阵的行数(row)相同时才有定义。一般单指矩阵乘积时,指的便是一般矩阵乘积。若A为i×r矩阵,B为r×j矩阵,则他们的乘积AB(有时记做A · B)会是一个i×j矩阵。其乘积矩阵的元素如下面式子得出:\
书中提到的对矩阵乘法的MapReduce实现方法是:
Map函数:对于矩阵M的每个元素M[i,j],产生一系列的键值对(i,k)->(M,j, M[i,j]),其中k=1,2…,直到矩阵N的列数。同样,对于矩阵N的每个元素N[j,k],产生一系列的键值对(i,k)->(N,j,N[j,k]),其中i=1,2…,直到矩阵M的行数。
Reduce函数:根据MR的原理,相同键i,k的数据会发送个同一个 reduce。如果M为2*2矩阵,N为2×3矩阵,reduce函数需要处理的数据为:
(1,1)->[(M,1, M[1,1])、(M,2, M[1,2])、(N,1, N[1,1])、(N,2, N[2,1])],
(1,2)->[(M,1, M[1,1])、(M,2, M[1,2])、(N,1, N[1,2])、(N,2, N[2,2])],
(1,3)->[(M,1, M[1,1])、(M,2, M[1,2])、(N,1, N[1,3])、(N,2, N[2,3])],
(2,1)->[(M,1, M[2,1])、(M,2, M[2,2])、(N,1, N[1,1])、(N,2, N[2,1])],
(2,2)->[(M,1, M[2,1])、(M,2, M[2,2])、(N,1, N[1,2])、(N,2, N[2,2])],
(2,3)->[(M,1, M[2,1])、(M,2, M[2,2])、(N,1, N[1,3])、(N,2, N[2,3])]。
这样只要将所有(M,j, M[i,j])和(N,j, N[j,k])分别按照j值排序并放在不同的两个列表里面。将这个列表的第j个元素M[i,j]个N[j,k]相乘,然后将这些积相加,最后积的和与键(i,k)组对作为reduce函数的输出。对于上面的例子reduce的输出就是:
(1,1)->(M[1,1]* N[1,1]+ M[1,2]* N[2,1])
(1,2)->(M[1,1]* N[1,2]+ M[1,2]* N[2,2])
(1,3)->(M[1,1]* N[1,3]+ M[1,2]* N[2,3])
(2,1)->(M[2,1]* N[2,1]+ M[2,2]* N[2,1])
(2,2)->(M[2,1]* N[1,2]+ M[2,2]* N[2,2])
(2,3)->(M[2,1]* N[1,3]+ M[2,2]* N[2,3])
下面是MapReduce的实现步骤:
(1).构造矩阵M:300*150;矩阵N:150*500。两矩阵的值放在一个M.data文件中,每行的格式为:文件标识#行坐标#列坐标#坐标值。
package com.cuijh.matrix;
import java.io.IOException;
import java.net.URI;
import java.util.Random;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataOutputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
//构造矩阵 M: 300*150 N: 150*300 两矩阵的值放在一个Matrix.data文件中,每行的格式为:文件标识#行坐标#列坐标#坐标值。
public class MatrixGenerate {
public static void main(String[] args) throws IOException, InterruptedException {
Configuration conf = new Configuration();
String dst = "hdfs://127.0.0.1:9000/user/cuijh/testdata/Matrix.data";
FileSystem fs = FileSystem.get(URI.create(dst), conf);
Path path = new Path(dst);
FSDataOutputStream fdos= fs.create(path,true);
int i=30; int j=15; int k=50;
for(int r=1; r<=i; r++){
for(int c=1; c<=j; c++){
fdos.writeUTF("M#"+r+"#"+c+"#"+new Random().nextInt(100)+"\n");
}
fdos.flush();
}
for(int r=1; r<=j; r++){
for(int c=1; c<=k; c++){
fdos.writeUTF("N#"+r+"#"+c+"#"+new Random().nextInt(100)+"\n");
}
fdos.flush();
}
fdos.close();
fs.close();
}
}
(2).基于上面的方法编写Map函数和Reduce函数。
package com.cuijh.matrix;
import java.io.IOException;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.util.GenericOptionsParser;
//单轮MapReduce的矩阵乘法
public class MartrixMultiplication {
public static class MartrixMapper extends Mapper<Object, Text, Text, Text>{
private Text map_key = new Text();
private Text map_value = new Text();
int rNumber = 30;
int cNumber = 50;
String fileTarget;
String i,j,k,ij,jk;
public void map(Object key, Text value, Context context) throws IOException, InterruptedException {
String eachterm[] = value.toString().split("#");
fileTarget = eachterm[0];
if(fileTarget.equals("M")){
i = eachterm[1];
j = eachterm[2];
ij = eachterm[3];
for(int c = 1; c<=cNumber; c++){
map_key.set(i + "#" + String.valueOf(c));
map_value.set("M" + "#" + j + "#" + ij);
context.write(map_key, map_value);
}
}else if(fileTarget.equals("N")){
j = eachterm[1];
k = eachterm[2];
jk = eachterm[3];
for(int r = 1; r<=rNumber; r++){
map_key.set(String.valueOf(r) + "#" +k);
map_value.set("N" + "#" + j + "#" + jk);
context.write(map_key, map_value);
}
}
}
}
public static class MartrixReducer extends Reducer<Text,Text,Text,Text> {
private Text reduce_value = new Text();
int jNumber = 15;
int M_ij[] = new int[jNumber+1];
int N_jk[] = new int[jNumber+1];
int j, ij, jk;
String fileTarget;
public void reduce(Text key, Iterable<Text> values, Context context) throws IOException, InterruptedException {
int jsum=0;
for(Text val : values){
String eachterm[] = val.toString().split("#");
fileTarget = eachterm[0];
j=Integer.parseInt(eachterm[1]);
if("M".equals(fileTarget)){
ij = Integer.parseInt(eachterm[2]);
M_ij[j]=ij;
}else if("N".equals(fileTarget)){
jk = Integer.parseInt(eachterm[2]);
N_jk[j]=jk;
}
}
for(int d=1; d<=jNumber; d++){
jsum += M_ij[j] * N_jk[d];
}
reduce_value.set(String.valueOf(jsum));
context.write(key, reduce_value);
}
}
public static void main(String[] args) throws IOException, InterruptedException, ClassNotFoundException {
Configuration conf = new Configuration();
String[] otherArgs = new GenericOptionsParser(conf,args).getRemainingArgs();
if(otherArgs.length !=2){
System.out.println("Usage: MatrixMultiplication <input> <output>");
System.exit(2);
}
Job job = new Job(conf,"MartrixMultiplication");
job.setJarByClass(MartrixMultiplication.class);
job.setMapperClass(MartrixMapper.class);
job.setReducerClass(MartrixReducer.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);
FileInputFormat.setInputPaths(job, new Path(otherArgs[0]));
FileOutputFormat.setOutputPath(job, new Path(otherArgs[1]));
System.exit(job.waitForCompletion(true) ? 0 : 1);
}
}