Mahout分布式程序开发 聚类Kmeans

前言
Mahout是基于Hadoop用于机器学习的程序开发框架，Mahout封装了3大类的机器学习算法，其中包括聚类算法。Kmeans是我们经常会提到的聚类算法之一，特别是处理未知数据集的时候，都会先聚类一下，看看数据集会有些什么样的规则。
本文主要讲解，基于Mahout程序开发，实现分布式的Kmeans算法。

---

目录

1. 聚类算法Kmeans
2. Mahout开发环境介绍
3. 用Mahout实现聚类算法Kmeans
4. 用R语言可视化结果
5. 模板项目上传github

---

1.聚类算法Kmeans

聚类分析是数据挖掘及机器学习领域内的重点问题之一，在数据挖掘、模式识别、决策支持、机器学习及图像分割等领域有广泛的应用，是最重要的数据分析方法之一。聚类是在给定的数据集合中寻找同类的数据子集合，每一个子集合形成一个类簇，同类簇中的数据具有更大的相似性。聚类算法大体上可分为基于划分的方法、基于层次的方法、基于密度的方法、基于网格的方法以及基于模型的方法。
k-means algorithm算法是一种得到最广泛使用的基于划分的聚类算法，把n个对象分为k个簇，以使簇内具有较高的相似度。相似度的计算根据一个簇中对象的平均值来进行。它与处理混合正态分布的最大期望算法很相似，因为他们都试图找到数据中自然聚类的中心。
算法首先随机地选择k个对象，每个对象初始地代表了一个簇的平均值或中心。对剩余的每个对象根据其与各个簇中心的距离，将它赋给最近的簇，然后重新计算每个簇的平均值。这个过程不断重复，直到准则函数收敛。

---

2.Mahout开发环境介绍

接上篇文章：Mahout分布式程序开发 基于物品的协同过滤ItemCF
所有环境变量和系统配置与上文一致！

---

3.用Mahout实现聚类算法Kmeans

实现步骤:

1. 准备数据文件: randomData.csv
2. Java程序：KmeansHadoop.java
3. 运行程序
4. 聚类结果解读
5. HDFS产生的目录

---

1).准备数据文件:randomData.csv
数据文件randomData.csv，由R语言通过“随机正太分布函数”程序生成，单机内存实验请参考文章：用Maven构建Mahout项目
原始数据文件：这里只截取一部分数据。

~ vi datafile/randomData.csv

-0.883033363823402 -3.31967192630249
-2.39312626419456 3.34726861118871
2.66976353341256 1.85144276077058
-1.09922906899594 -6.06261735207489
-4.36361936997216 1.90509905380532
-0.00351835125495037 -0.610105996559153
-2.9962958796338 -3.60959839525735
-3.27529418132066 0.0230099799641799
2.17665594420569 6.77290756817957
-2.47862038335637 2.53431833167278
5.53654901906814 2.65089785582474
5.66257474538338 6.86783609641077
-0.558946883114376 1.22332819416237
5.11728525486132 3.74663871584768
1.91240516693351 2.95874731384062
-2.49747101306535 2.05006504756875
3.98781883213459 1.00780938946366
5.47470532716682 5.35084411045171

注：由于Mahout中kmeans算法，默认的分隔符是空格，因此我把逗号分隔的数据文件，改为以空格为分隔的。
2).Java程序：KmeansHadoop.java
Kmeans的算法实现，请查看Mahout in Action。

package org.conan.mymahout.cluster08;

import org.apache.hadoop.fs.Path;
import org.apache.hadoop.mapred.JobConf;
import org.apache.mahout.clustering.conversion.InputDriver;
import org.apache.mahout.clustering.kmeans.KMeansDriver;
import org.apache.mahout.clustering.kmeans.RandomSeedGenerator;
import org.apache.mahout.common.distance.DistanceMeasure;
import org.apache.mahout.common.distance.EuclideanDistanceMeasure;
import org.apache.mahout.utils.clustering.ClusterDumper;
import org.conan.mymahout.hdfs.HdfsDAO;
import org.conan.mymahout.recommendation.ItemCFHadoop;

public class KmeansHadoop {
    private static final String HDFS = "hdfs://192.168.1.210:9000";

    public static void main(String[] args) throws Exception {
        String localFile = "datafile/randomData.csv";
        String inPath = HDFS + "/user/hdfs/mix_data";
        String seqFile = inPath + "/seqfile";
        String seeds = inPath + "/seeds";
        String outPath = inPath + "/result/";
        String clusteredPoints = outPath + "/clusteredPoints";

        JobConf conf = config();
        HdfsDAO hdfs = new HdfsDAO(HDFS, conf);
        hdfs.rmr(inPath);
        hdfs.mkdirs(inPath);
        hdfs.copyFile(localFile, inPath);
        hdfs.ls(inPath);

        InputDriver.runJob(new Path(inPath), new Path(seqFile), "org.apache.mahout.math.RandomAccessSparseVector");

        int k = 3;
        Path seqFilePath = new Path(seqFile);
        Path clustersSeeds = new Path(seeds);
        DistanceMeasure measure = new EuclideanDistanceMeasure();
        clustersSeeds = RandomSeedGenerator.buildRandom(conf, seqFilePath, clustersSeeds, k, measure);
        KMeansDriver.run(conf, seqFilePath, clustersSeeds, new Path(outPath), measure, 0.01, 10, true, 0.01, false);

        Path outGlobPath = new Path(outPath, "clusters-*-final");
        Path clusteredPointsPath = new Path(clusteredPoints);
        System.out.printf("Dumping out clusters from clusters: %s and clusteredPoints: %s\n", outGlobPath, clusteredPointsPath);

        ClusterDumper clusterDumper = new ClusterDumper(outGlobPath, clusteredPointsPath);
        clusterDumper.printClusters(null);
    }

    public static JobConf config() {
        JobConf conf = new JobConf(ItemCFHadoop.class);
        conf.setJobName("ItemCFHadoop");
        conf.addResource("classpath:/hadoop/core-site.xml");
        conf.addResource("classpath:/hadoop/hdfs-site.xml");
        conf.addResource("classpath:/hadoop/mapred-site.xml");
        return conf;
    }

}

3).运行程序
控制台输出：

Delete: hdfs://192.168.1.210:9000/user/hdfs/mix_data
Create: hdfs://192.168.1.210:9000/user/hdfs/mix_data
copy from: datafile/randomData.csv to hdfs://192.168.1.210:9000/user/hdfs/mix_data
ls: hdfs://192.168.1.210:9000/user/hdfs/mix_data
==========================================================
name: hdfs://192.168.1.210:9000/user/hdfs/mix_data/randomData.csv, folder: false, size: 36655
==========================================================
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
2013-10-14 15:39:31 org.apache.hadoop.util.NativeCodeLoader 
警告: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
2013-10-14 15:39:31 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:31 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:31 org.apache.hadoop.io.compress.snappy.LoadSnappy 
警告: Snappy native library not loaded
2013-10-14 15:39:31 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0001
2013-10-14 15:39:31 org.apache.hadoop.mapred.Task initialize
信息:  Using ResourceCalculatorPlugin : null
2013-10-14 15:39:31 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0001_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:31 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:31 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0001_m_000000_0 is allowed to commit now
2013-10-14 15:39:31 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0001_m_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/seqfile
2013-10-14 15:39:31 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:31 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0001_m_000000_0' done.
2013-10-14 15:39:32 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息:  map 100% reduce 0%
2013-10-14 15:39:32 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0001
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息: Counters: 11
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:   File Output Format Counters 
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Bytes Written=31390
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:   File Input Format Counters 
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Bytes Read=36655
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:   FileSystemCounters
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     FILE_BYTES_READ=475910
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     HDFS_BYTES_READ=36655
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     FILE_BYTES_WRITTEN=506350
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     HDFS_BYTES_WRITTEN=68045
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:   Map-Reduce Framework
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Map input records=1000
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Spilled Records=0
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Total committed heap usage (bytes)=188284928
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     SPLIT_RAW_BYTES=124
2013-10-14 15:39:32 org.apache.hadoop.mapred.Counters log
信息:     Map output records=1000
2013-10-14 15:39:32 org.apache.hadoop.io.compress.CodecPool getCompressor
信息: Got brand-new compressor
2013-10-14 15:39:32 org.apache.hadoop.io.compress.CodecPool getDecompressor
信息: Got brand-new decompressor
2013-10-14 15:39:32 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:32 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:32 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0002
2013-10-14 15:39:32 org.apache.hadoop.mapred.Task initialize
信息:  Using ResourceCalculatorPlugin : null
2013-10-14 15:39:32 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: io.sort.mb = 100
2013-10-14 15:39:32 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: data buffer = 79691776/99614720
2013-10-14 15:39:32 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: record buffer = 262144/327680
2013-10-14 15:39:33 org.apache.hadoop.mapred.MapTask$MapOutputBuffer flush
信息: Starting flush of map output
2013-10-14 15:39:33 org.apache.hadoop.mapred.MapTask$MapOutputBuffer sortAndSpill
信息: Finished spill 0
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0002_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:33 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0002_m_000000_0' done.
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task initialize
信息:  Using ResourceCalculatorPlugin : null
2013-10-14 15:39:33 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Merging 1 sorted segments
2013-10-14 15:39:33 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Down to the last merge-pass, with 1 segments left of total size: 623 bytes
2013-10-14 15:39:33 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0002_r_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:33 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0002_r_000000_0 is allowed to commit now
2013-10-14 15:39:33 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0002_r_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-1
2013-10-14 15:39:33 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: reduce > reduce
2013-10-14 15:39:33 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0002_r_000000_0' done.
2013-10-14 15:39:33 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息:  map 100% reduce 100%
2013-10-14 15:39:33 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0002
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息: Counters: 19
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:   File Output Format Counters 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Bytes Written=695
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:   FileSystemCounters
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     FILE_BYTES_READ=4239303
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     HDFS_BYTES_READ=203963
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     FILE_BYTES_WRITTEN=4457168
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     HDFS_BYTES_WRITTEN=140321
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:   File Input Format Counters 
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Bytes Read=31390
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:   Map-Reduce Framework
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Map output materialized bytes=627
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Map input records=1000
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Reduce shuffle bytes=0
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Spilled Records=6
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Map output bytes=612
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Total committed heap usage (bytes)=376569856
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     SPLIT_RAW_BYTES=130
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Combine input records=0
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Reduce input records=3
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Reduce input groups=3
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Combine output records=0
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Reduce output records=3
2013-10-14 15:39:33 org.apache.hadoop.mapred.Counters log
信息:     Map output records=3
2013-10-14 15:39:34 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:34 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:34 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0003
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task initialize
信息:  Using ResourceCalculatorPlugin : null
2013-10-14 15:39:34 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: io.sort.mb = 100
2013-10-14 15:39:34 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: data buffer = 79691776/99614720
2013-10-14 15:39:34 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: record buffer = 262144/327680
2013-10-14 15:39:34 org.apache.hadoop.mapred.MapTask$MapOutputBuffer flush
信息: Starting flush of map output
2013-10-14 15:39:34 org.apache.hadoop.mapred.MapTask$MapOutputBuffer sortAndSpill
信息: Finished spill 0
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0003_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:34 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0003_m_000000_0' done.
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task initialize
信息:  Using ResourceCalculatorPlugin : null
2013-10-14 15:39:34 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:34 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Merging 1 sorted segments
2013-10-14 15:39:34 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Down to the last merge-pass, with 1 segments left of total size: 677 bytes
2013-10-14 15:39:34 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0003_r_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:34 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:34 org.apache.hadoop.mapred.Task commit