hadoop2.2编程:矩阵相乘简单实现
/* matrix-matrix multiplication on Hadoop A x B = C constraint: A, B, C must be of the same size I use this to evaluate the efficiency of Hadoop for matrix multiplication, so I really don't care to handle non-square matrices. ===Data preparation==== Matrix data must be stored in a file on Hadoop. Line number must be appended to the beginning of each line. For example, the following represents a 4x4 matrix: 0 18 20 16 14 1 17 12 11 19 2 10 17 11 19 3 14 17 20 10 Left (A in this example) matrix should be stored in file "left"; Right (B in this example) matrix should be stored in file "right"; I use filenames to distinguish input data. Place "left" and "right" in the same folder (let's call it "input") ====Run the program==== > hadoop jar matrixmul.jar MatrixMul input output 8 2 results will be placed in "output" folder on HDFS. 8: all matrices are 8x8 2: every partitioned block is of size 2x2 ===Read the results=== Given the above sample command, we multiply two 8x8 matrices, in many 2x2 blocks. So, that the resulted C matrix has 16 blocks. In the output folder, there will be 16 separate files: part-r-00000, part-r-00001, ... part-r-00015 Every file stores one block in C. In this example, every block has 2 rows and 2 columns. These files are organized in "row"-order. ===Algorithm=== Mappers read input data. Every reducer processes one block of the resulted matrix. */ import java.io.IOException; import java.util.StringTokenizer; import org.apache.hadoop.conf.Configuration; import org.apache.hadoop.fs.Path; import org.apache.hadoop.io.IntWritable; import org.apache.hadoop.io.LongWritable; 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.input.TextInputFormat; import org.apache.hadoop.mapreduce.lib.input.FileSplit; import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat; import org.apache.hadoop.util.GenericOptionsParser; public class MatrixMul { public static class MyMapper extends Mapper<LongWritable, Text, IntWritable, Text>{ private String filename=null; private boolean isLeftMatrix=false; private int totalSize, partSize, npart; private boolean isLeft(){return isLeftMatrix;} protected void setup(Context context) throws IOException, InterruptedException{ //get filename FileSplit fileSplit = (FileSplit)context.getInputSplit(); filename = fileSplit.getPath().getName(); if("left".equalsIgnoreCase(filename)) isLeftMatrix=true; else isLeftMatrix=false; //get how size and partition information Configuration conf=context.getConfiguration(); totalSize=conf.getInt("matrix-mul-totalsize", -1); partSize=conf.getInt("matrix-mul-partsize", -1); npart=conf.getInt("matrix-mul-npart", -1); if(totalSize<0 || partSize<0 || npart<0){ System.out.println("Error in setup of MyMapper."); System.exit(1); } } public void map(LongWritable key, Text value, Context context ) throws IOException, InterruptedException { String line=value.toString(); String[] strs=line.split(" "); if(strs.length!=totalSize+1){ System.out.println("Error in map of Mapper."); System.out.println(strs.length+"___"+totalSize); System.out.println("line is: "+line); System.exit(1); } int linenum=Integer.parseInt(strs[0]); int[] numbers=new int[totalSize]; for(int i=0;i<totalSize;i++) numbers[i]=Integer.parseInt(strs[i+1]); int part_hor=linenum/partSize; //horizontal partitioned id int prev_part_ver=-1; String msg=null; for(int i=0;i<totalSize;i++){ int part_ver=i/partSize; //vertical partition number if(part_ver!=prev_part_ver){ if(msg!=null){ int baselinenum = part_hor * partSize; int old=part_ver; part_ver=prev_part_ver; if(isLeft()){ String toSend="l:"+(linenum - baselinenum)+":"+part_ver+"#"+msg; System.out.println("left "+linenum+","+part_ver+" "+msg); for(int k=0;k<npart;k++){ int dest=part_hor * npart + k; context.write(new IntWritable(dest), new Text(toSend)); } }else{ String toSend="r:"+(linenum - baselinenum)+":"+part_hor+"#"+msg; System.out.println("right "+part_ver+":"+linenum+" "+msg); for(int k=0;k<npart;k++){ int dest=k * npart + part_ver; context.write(new IntWritable(dest), new Text(toSend)); } } part_ver=old; } msg=null; prev_part_ver=part_ver; } if(msg==null) msg=""+strs[i+1]; else msg+=" "+strs[i+1]; } if(msg!=null){ //almost the same code int part_ver=npart-1; int baselinenum = part_hor * partSize; if(isLeft()){ String toSend="l:"+(linenum - baselinenum)+":"+part_ver+"#"+msg; System.out.println("left "+linenum+","+part_ver+" "+msg); for(int k=0;k<npart;k++){ int dest=part_hor * npart + k; context.write(new IntWritable(dest), new Text(toSend)); } }else{ String toSend="r:"+(linenum - baselinenum)+":"+part_hor+"#"+msg; System.out.println("right "+part_ver+":"+linenum+" "+msg); for(int k=0;k<npart;k++){ int dest=k * npart + part_ver; //has to be the last part context.write(new IntWritable(dest), new Text(toSend)); } } } } } public static class MyReducer extends Reducer<IntWritable, Text, Text, Text> { private int totalSize, partSize, npart; int[][] left=null; int[][] right=null; protected void setup(Context context) throws IOException, InterruptedException{ //get how # of partitions Configuration conf=context.getConfiguration(); totalSize=conf.getInt("matrix-mul-totalsize", -1); partSize=conf.getInt("matrix-mul-partsize", -1); npart=conf.getInt("matrix-mul-npart", -1); if(totalSize<0 || partSize<0 || npart<0){ System.out.println("Error in setup of MyReducer."); System.exit(1); } left=new int[partSize][totalSize]; right=new int[totalSize][partSize]; } public void reduce(IntWritable key, Iterable<Text> values, Context context ) throws IOException, InterruptedException { int sum = 0; for (Text val : values) { String line=val.toString(); String[] meta_val=line.split("#"); String[] metas=meta_val[0].split(":"); String[] numbers=meta_val[1].split(" "); int baselinenum=Integer.parseInt(metas[1]); int blkindex=Integer.parseInt(metas[2]); if("l".equalsIgnoreCase(metas[0])){ //from left matrix int start=blkindex * partSize; for(int i=0;i<partSize; i++) left[baselinenum][start+i]=Integer.parseInt(numbers[i]); }else{ int rowindex=blkindex*partSize + baselinenum; for(int i=0;i<partSize; i++) right[rowindex][i]=Integer.parseInt(numbers[i]); } } } protected void cleanup(Context context) throws IOException, InterruptedException { //now let's do the calculation int[][] res=new int[partSize][partSize]; for(int i=0;i<partSize;i++) for(int j=0;j<partSize;j++) res[i][j]=0; for(int i=0;i<partSize;i++){ for(int k=0;k<totalSize;k++){ for(int j=0;j<partSize;j++){ res[i][j]+=left[i][k]*right[k][j]; } } } for(int i=0;i<partSize;i++){ String output=null; for(int j=0;j<partSize;j++){ if(output==null) output=""+res[i][j]; else output+=" "+res[i][j]; } context.write(new Text(output), null); } } } public static void main(String[] args) throws Exception { Configuration conf = new Configuration(); if (args.length != 4) { System.err.println("Usage: MatrixMul input-dir output-dir total-size part-size"); System.exit(2); } int totalsize=Integer.parseInt(args[2]); int partsize=Integer.parseInt(args[3]); if(totalsize==0 || partsize==0 || partsize>totalsize){ System.out.println("Invalid total-size or part-size"); System.exit(1); } conf.setInt("matrix-mul-totalsize", totalsize); //the matrix is 'totalsize' by 'totalsize' conf.setInt("matrix-mul-partsize", partsize); //every block is 'partsize' by 'partsize' int npart=totalsize/partsize; if(npart*partsize<totalsize) npart++; conf.setInt("matrix-mul-npart", npart); //number of parts on one dimension Job job = new Job(conf, "matrix-mul"); job.setJarByClass(MatrixMul.class); job.setMapperClass(MyMapper.class); job.setReducerClass(MyReducer.class); job.setNumReduceTasks(npart*npart); job.setOutputKeyClass(IntWritable.class); job.setOutputValueClass(Text.class); //FileInputFormat.addInputPath(job, new Path(args[0])); TextInputFormat.addInputPath(job, new Path(args[0])); //need to read a complete line FileOutputFormat.setOutputPath(job, new Path(args[1])); job.waitForCompletion(true) ; } }
hadoop2.2编程:矩阵相乘简单实现的更多相关文章
- CUDA编程-(2)其实写个矩阵相乘并不是那么难
程序代码及图解析: #include <iostream> #include "book.h" __global__ void add( int a, int b, i ...
- 编程计算2×3阶矩阵A和3×2阶矩阵B之积C。 矩阵相乘的基本方法是: 矩阵A的第i行的所有元素同矩阵B第j列的元素对应相乘, 并把相乘的结果相加,最终得到的值就是矩阵C的第i行第j列的值。 要求: (1)从键盘分别输入矩阵A和B, 输出乘积矩阵C (2) **输入提示信息为: 输入矩阵A之前提示:"Input 2*3 matrix a:\n" 输入矩阵B之前提示
编程计算2×3阶矩阵A和3×2阶矩阵B之积C. 矩阵相乘的基本方法是: 矩阵A的第i行的所有元素同矩阵B第j列的元素对应相乘, 并把相乘的结果相加,最终得到的值就是矩阵C的第i行第j列的值. 要求: ...
- 利用Hadoop实现超大矩阵相乘之我见(二)
前文 在<利用Hadoop实现超大矩阵相乘之我见(一)>中我们所介绍的方法有着“计算过程中文件占用存储空间大”这个缺陷,本文中我们着重解决这个问题. 矩阵相乘计算思想 传统的矩阵相乘方法为 ...
- 利用Hadoop实现超大矩阵相乘之我见(一)
前记 最近,公司一位挺优秀的总务离职,欢送宴上,她对我说“你是一位挺优秀的程序员”,刚说完,立马道歉说“对不起,我说你是程序员是不是侮辱你了?”我挺诧异,程序员现在是很低端,很被人瞧不起的工作吗?或许 ...
- Strassen 矩阵相乘算法(转)
偶尔在算法课本上面看到矩阵相乘的算法,联想到自己曾经在蓝桥杯系统上曾经做过一道矩阵相乘的题目,当时用的是普通的矩阵相乘的方法,效率极低,勉强通过编译.所以决定研究一下Strassen矩阵相乘算法,由于 ...
- dp方法论——由矩阵相乘问题学习dp解题思路
前篇戳:dp入门——由分杆问题认识动态规划 导语 刷过一些算法题,就会十分珍惜“方法论”这种东西.Leetcode上只有题目.讨论和答案,没有方法论.往往答案看起来十分切中要害,但是从看题目到得到思路 ...
- Opencv中Mat矩阵相乘——点乘、dot、mul运算详解
Opencv中Mat矩阵相乘——点乘.dot.mul运算详解 2016年09月02日 00:00:36 -牧野- 阅读数:59593 标签: Opencv矩阵相乘点乘dotmul 更多 个人分类: O ...
- C++两个矩阵相乘
/*编程求两个矩阵相乘的结果.输入第一行是整数m,n,表示第一个矩阵式m行n列的:然后是一个m * n的矩阵.再下一行的输入时整数p,q,表示下一个矩阵p行,q列的(n=p);然后就是一个p行q列的矩 ...
- 使用cublas 矩阵库函数实现矩阵相乘
2014-08-10 cublas中执行矩阵乘法运算的函数 首先要注意的是cublas使用的是以列为主的存储方式,和c/c++中的以行为主的方式是不一样的.处理方法可参考下面的注释代码 // SOME ...
随机推荐
- Poj 1001 / OpenJudge 2951 Exponentiation
1.链接地址: http://poj.org/problem?id=1001 http://bailian.openjudge.cn/practice/2951 2.题目: Exponentiatio ...
- Linux下UDP收/发广播消息简单实现
发送广播消息 #include<stdio.h> #include<stdlib.h> #include<string.h> #include<sys/typ ...
- Putty终端 模拟 远程登录 虚拟机Linux
1.虚拟机设置 虚拟机设置->网络适配器->选择Host-only:与主机共享一个私有网络 桥接.NAT.Host-only三种网络模式的说明: (1)桥接:表示在局域网内是一台真实的系统 ...
- outlet删除不完全
今天在用iOS写个计算器的时候,遇到的一个小bug,新手,太新了,不之所错. 直接上码: Terminating app due to uncaught exception 'NSUnknownKey ...
- 《APUE》第四章笔记(2)
下面介绍对stat结构的各个成员的操作函数. 先贴个stat结构的图: access函数: #include <unistd.h> int access(const char *pathn ...
- [CSS]font- 属性
所有浏览器都支持 font 属性. 注释:任何的版本的 Internet Explorer (包括 IE8)都不支持属性值 "inherit". 定义和用法 font 简写属性在一 ...
- 关于SQL优化的一个小试例子
原SQL: select ta.serialno, ta.accepttime, ta.subsnumber, ta.subsname, ta.cont ...
- c语言贪吃蛇
思路:函数gotoxy(x,y)使光标移植屏幕的x,y坐标(屏幕左上角为0,0),用来绘制蛇和界面,color()函数用来设置绘制的颜色.设有snakelong节,第i节蛇的x坐标为x[i],y坐标为 ...
- linux系统管理
系统 # uname -a # 查看内核/操作系统/CPU信息# head -n 1 /etc/issue # 查看操作系统版本# cat /proc/cpuinfo ...
- Java 内存区域
Java虚拟机锁管理的内存包括以下几个运行时数据区域: 1.程序计数器: 程序计数器是一块较小的内存空间,它的作用可以看做是当前线程所执行的字节码的行号指示器.在虚拟机的概念模型里,字节码解释器工作时 ...