标签：mahout   mapreduce   hadoop   机器学习   数据挖掘   

1. 聚类算法kmeans

2. Mahout开发环境介绍

3. 用Mahout实现聚类算法kmeans

• 准备数据文件: randomData.csv
• Java程序：KmeansHadoop.java
• 运行程序
• 聚类结果解读
• 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：这里只截取了一部分数据。

~ 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，默认的分融符是” “(空格)，因些我把逗号分隔的数据文件，改成以空格分隔。

2). Java程序：KmeansHadoop.javakmeans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 15:39:41 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Down to the last merge-pass, with 1 segments left of total size: 677 bytes
2013-10-14 15:39:41 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:41 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0009_r_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:41 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:41 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0009_r_000000_0 is allowed to commit now
2013-10-14 15:39:41 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0009_r_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-8
2013-10-14 15:39:41 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: reduce > reduce
2013-10-14 15:39:41 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0009_r_000000_0' done.
2013-10-14 15:39:42 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: map 100% reduce 100%
2013-10-14 15:39:42 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0009
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Counters: 19
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: File Output Format Counters 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Bytes Written=695
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: FileSystemCounters
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_READ=27256775
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_READ=673669
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_WRITTEN=28569192
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_WRITTEN=152767
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: File Input Format Counters 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Bytes Read=31390
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Map-Reduce Framework
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Map output materialized bytes=681
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Map input records=1000
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Reduce shuffle bytes=0
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Spilled Records=6
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Map output bytes=666
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Total committed heap usage (bytes)=1772093440
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: SPLIT_RAW_BYTES=130
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Combine input records=0
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Reduce input records=3
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Reduce input groups=3
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Combine output records=0
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Reduce output records=3
2013-10-14 15:39:42 org.apache.hadoop.mapred.Counters log
信息: Map output records=3
2013-10-14 15:39:42 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:42 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:42 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0010
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task initialize
信息: Using ResourceCalculatorPlugin : null
2013-10-14 15:39:42 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: io.sort.mb = 100
2013-10-14 15:39:42 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: data buffer = 79691776/99614720
2013-10-14 15:39:42 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: record buffer = 262144/327680
2013-10-14 15:39:42 org.apache.hadoop.mapred.MapTask$MapOutputBuffer flush
信息: Starting flush of map output
2013-10-14 15:39:42 org.apache.hadoop.mapred.MapTask$MapOutputBuffer sortAndSpill
信息: Finished spill 0
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0010_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:42 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0010_m_000000_0' done.
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task initialize
信息: Using ResourceCalculatorPlugin : null
2013-10-14 15:39:42 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Merging 1 sorted segments
2013-10-14 15:39:42 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Down to the last merge-pass, with 1 segments left of total size: 677 bytes
2013-10-14 15:39:42 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0010_r_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:42 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0010_r_000000_0 is allowed to commit now
2013-10-14 15:39:42 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0010_r_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-9
2013-10-14 15:39:42 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: reduce > reduce
2013-10-14 15:39:42 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0010_r_000000_0' done.
2013-10-14 15:39:43 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: map 100% reduce 100%
2013-10-14 15:39:43 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0010
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Counters: 19
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: File Output Format Counters 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Bytes Written=695
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: FileSystemCounters
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_READ=30544993
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_READ=741007
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_WRITTEN=32013760
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_WRITTEN=154545
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: File Input Format Counters 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Bytes Read=31390
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Map-Reduce Framework
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Map output materialized bytes=681
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Map input records=1000
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Reduce shuffle bytes=0
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Spilled Records=6
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Map output bytes=666
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Total committed heap usage (bytes)=1966735360
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: SPLIT_RAW_BYTES=130
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Combine input records=0
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Reduce input records=3
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Reduce input groups=3
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Combine output records=0
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Reduce output records=3
2013-10-14 15:39:43 org.apache.hadoop.mapred.Counters log
信息: Map output records=3
2013-10-14 15:39:43 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:43 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:43 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0011
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task initialize
信息: Using ResourceCalculatorPlugin : null
2013-10-14 15:39:43 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: io.sort.mb = 100
2013-10-14 15:39:43 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: data buffer = 79691776/99614720
2013-10-14 15:39:43 org.apache.hadoop.mapred.MapTask$MapOutputBuffer 
信息: record buffer = 262144/327680
2013-10-14 15:39:43 org.apache.hadoop.mapred.MapTask$MapOutputBuffer flush
信息: Starting flush of map output
2013-10-14 15:39:43 org.apache.hadoop.mapred.MapTask$MapOutputBuffer sortAndSpill
信息: Finished spill 0
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0011_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:43 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0011_m_000000_0' done.
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task initialize
信息: Using ResourceCalculatorPlugin : null
2013-10-14 15:39:43 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Merging 1 sorted segments
2013-10-14 15:39:43 org.apache.hadoop.mapred.Merger$MergeQueue merge
信息: Down to the last merge-pass, with 1 segments left of total size: 677 bytes
2013-10-14 15:39:43 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0011_r_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:43 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0011_r_000000_0 is allowed to commit now
2013-10-14 15:39:43 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0011_r_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-10
2013-10-14 15:39:43 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: reduce > reduce
2013-10-14 15:39:43 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0011_r_000000_0' done.
2013-10-14 15:39:44 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: map 100% reduce 100%
2013-10-14 15:39:44 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0011
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Counters: 19
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: File Output Format Counters 
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Bytes Written=695
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: FileSystemCounters
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_READ=33833211
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_READ=808345
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_WRITTEN=35458320
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_WRITTEN=156323
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: File Input Format Counters 
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Bytes Read=31390
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Map-Reduce Framework
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Map output materialized bytes=681
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Map input records=1000
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Reduce shuffle bytes=0
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Spilled Records=6
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Map output bytes=666
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Total committed heap usage (bytes)=2166095872
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: SPLIT_RAW_BYTES=130
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Combine input records=0
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Reduce input records=3
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Reduce input groups=3
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Combine output records=0
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Reduce output records=3
2013-10-14 15:39:44 org.apache.hadoop.mapred.Counters log
信息: Map output records=3
2013-10-14 15:39:44 org.apache.hadoop.mapred.JobClient copyAndConfigureFiles
警告: Use GenericOptionsParser for parsing the arguments. Applications should implement Tool for the same.
2013-10-14 15:39:44 org.apache.hadoop.mapreduce.lib.input.FileInputFormat listStatus
信息: Total input paths to process : 1
2013-10-14 15:39:44 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Running job: job_local_0012
2013-10-14 15:39:44 org.apache.hadoop.mapred.Task initialize
信息: Using ResourceCalculatorPlugin : null
2013-10-14 15:39:44 org.apache.hadoop.mapred.Task done
信息: Task:attempt_local_0012_m_000000_0 is done. And is in the process of commiting
2013-10-14 15:39:44 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:44 org.apache.hadoop.mapred.Task commit
信息: Task attempt_local_0012_m_000000_0 is allowed to commit now
2013-10-14 15:39:44 org.apache.hadoop.mapreduce.lib.output.FileOutputCommitter commitTask
信息: Saved output of task 'attempt_local_0012_m_000000_0' to hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusteredPoints
2013-10-14 15:39:44 org.apache.hadoop.mapred.LocalJobRunner$Job statusUpdate
信息: 
2013-10-14 15:39:44 org.apache.hadoop.mapred.Task sendDone
信息: Task 'attempt_local_0012_m_000000_0' done.
2013-10-14 15:39:45 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: map 100% reduce 0%
2013-10-14 15:39:45 org.apache.hadoop.mapred.JobClient monitorAndPrintJob
信息: Job complete: job_local_0012
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Counters: 11
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: File Output Format Counters 
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Bytes Written=41520
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: File Input Format Counters 
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Bytes Read=31390
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: FileSystemCounters
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_READ=18560374
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_READ=437203
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: FILE_BYTES_WRITTEN=19450325
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: HDFS_BYTES_WRITTEN=120417
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Map-Reduce Framework
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Map input records=1000
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Spilled Records=0
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Total committed heap usage (bytes)=1083047936
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: SPLIT_RAW_BYTES=130
2013-10-14 15:39:45 org.apache.hadoop.mapred.Counters log
信息: Map output records=1000
Dumping out clusters from clusters: hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-*-final and clusteredPoints: hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusteredPoints
CL-552{n=443 c=[1.631, -0.412] r=[1.563, 1.407]}
Weight : [props - optional]: Point:
1.0: [-2.393, 3.347]
1.0: [-4.364, 1.905]
1.0: [-3.275, 0.023]
1.0: [-2.479, 2.534]
1.0: [-0.559, 1.223]
...

CL-847{n=77 c=[-2.953, -0.971] r=[1.767, 2.189]}
Weight : [props - optional]: Point:
1.0: [-0.883, -3.320]
1.0: [-1.099, -6.063]
1.0: [-0.004, -0.610]
1.0: [-2.996, -3.610]
1.0: [3.988, 1.008]
...

CL-823{n=480 c=[0.219, 2.600] r=[1.479, 1.385]}
Weight : [props - optional]: Point:
1.0: [2.670, 1.851]
1.0: [2.177, 6.773]
1.0: [5.537, 2.651]
1.0: [5.663, 6.868]
1.0: [5.117, 3.747]
1.0: [1.912, 2.959]
...

4). 聚类结果解读

• 初始化环境
• 算法执行
• 打印聚类结果
  

HDFS的数据目录和工作目录，并上传数据文件。

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

b. 算法执行

• 把原始数据randomData.csv，转成Mahout sequence files of VectorWritable。
• 通过随机的方法，选中kmeans的3个中心，做为初始集群 。
• 根据迭代次数的设置，执行MapReduce，进行计算。
  

1):把原始数据randomData.csv，转成Mahout sequence files of VectorWritable。  程序源代码:

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

2):通过随机的方法，选中kmeans的3个中心，做为初始集群 程序源代码:

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);

日志输出：

Job complete: job_local_0002

3):根据迭代次数的设置，执行MapReduce，进行计算

程序源代码:

KMeansDriver.run(conf, seqFilePath, clustersSeeds, new Path(outPath), measure, 0.01, 10, true, 0.01, false);

日志输出：

Job complete: job_local_0003
Job complete: job_local_0004
Job complete: job_local_0005
Job complete: job_local_0006
Job complete: job_local_0007
Job complete: job_local_0008
Job complete: job_local_0009
Job complete: job_local_0010
Job complete: job_local_0011
Job complete: job_local_0012

c.打印聚类结果

Dumping out clusters from clusters: hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusters-*-final and clusteredPoints: hdfs://192.168.1.210:9000/user/hdfs/mix_data/result/clusteredPoints
CL-552{n=443 c=[1.631, -0.412] r=[1.563, 1.407]}
CL-847{n=77 c=[-2.953, -0.971] r=[1.767, 2.189]}
CL-823{n=480 c=[0.219, 2.600] r=[1.479, 1.385]}

运行结果：有3个中心。

Cluster1, 包括443个点，中心坐标[1.631, -0.412]
Cluster2, 包括77个点，中心坐标[-2.953, -0.971]
Cluster3, 包括480 个点，中心坐标[0.219, 2.600]

5). HDFS产生的目录

# 根目录
~ hadoop fs -ls /user/hdfs/mix_data
Found 4 items
-rw-r--r-- 3 Administrator supergroup 36655 2013-10-04 15:31 /user/hdfs/mix_data/randomData.csv
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/seeds
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/seqfile

# 输出目录
~ hadoop fs -ls /user/hdfs/mix_data/result
Found 13 items
-rw-r--r-- 3 Administrator supergroup 194 2013-10-04 15:31 /user/hdfs/mix_data/result/_policy
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusteredPoints
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-0
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-1
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-10-final
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-2
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-3
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-4
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-5
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-6
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-7
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-8
drwxr-xr-x - Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/result/clusters-9

# 产生的随机中心种子目录
~ hadoop fs -ls /user/hdfs/mix_data/seeds
Found 1 items
-rw-r--r-- 3 Administrator supergroup 599 2013-10-04 15:31 /user/hdfs/mix_data/seeds/part-randomSeed

# 输入文件换成Mahout格式文件的目录
~ hadoop fs -ls /user/hdfs/mix_data/seqfile
Found 2 items
-rw-r--r-- 3 Administrator supergroup 0 2013-10-04 15:31 /user/hdfs/mix_data/seqfile/_SUCCESS
-rw-r--r-- 3 Administrator supergroup 31390 2013-10-04 15:31 /user/hdfs/mix_data/seqfile/part-m-00000

4. 用R语言可视化结果

  分别把聚类后的点，保存到不同的cluster*.csv文件，然后用R语言画图。

c1<-read.csv(file="cluster1.csv",sep=",",header=FALSE)
c2<-read.csv(file="cluster2.csv",sep=",",header=FALSE)
c3<-read.csv(file="cluster3.csv",sep=",",header=FALSE)
y<-rbind(c1,c2,c3)
cols<-c(rep(1,nrow(c1)),rep(2,nrow(c2)),rep(3,nrow(c3)))
plot(y, col=c("black","blue","green")[cols])
center<-matrix(c(1.631, -0.412,-2.953, -0.971,0.219, 2.600),ncol=2,byrow=TRUE)
points(center, col="violetred", pch = 19)

  从上图中，我们看到有 黑，蓝，绿，三种颜色的空心点，这些点就是原始数据。3个紫色实点，是Mahout的kmeans后生成的3个中心。

Mahout分步式程序开发 聚类Kmeans

标签：mahout   mapreduce   hadoop   机器学习   数据挖掘   

原文地址：http://blog.csdn.net/u013361361/article/details/40558411