今天调研了并行化频繁模式挖掘算法PFP Growth及其在Mahout下的命令使用,简单记录下试验结果,供以后查阅:
- 环境:Jdk1.7 + Hadoop2.2.0单机伪集群 + Mahout0.6(0.8和0.9版本号都不包括该算法。Mahout0.6能够和Hadoop2.2.0和平共处有点意外orz)
- 部分输入数据,输入数据一行代表一个购物篮:
4750,19394,25651,6395,5592
26180,10895,24571,23295,20578,27791,2729,8637
7380,18805,25086,19048,3190,21995,10908,12576
3458,12426,20578
1880,10702,1731,5185,18575,28967
21815,10872,18730
20626,17921,28930,14580,2891,11080
18075,6548,28759,17133
7868,15200,13494
7868,28617,18097,22999,16323,8637,7045,25733
12189,8816,22950,18465,13258,27791,20979
26728
17512,14821,18741
26619,14470,21899,6731
5184
28653,28662,18353,27437,5661,12078,11849,15784,7248,7061,18612,24277,4807,15584,9671,18741,3647,1000
。。
。
。
。
。
- 运行命令:
mahout fpg -i /workspace/dataguru/hadoopdev/week13/fpg/in/ -o /workspace/dataguru/hadoopdev/week13/fpg/out -method mapreduce -s 3
參数说明:
-i 输入路径,因为执行在hadoop环境中,所以输入路径必须是hdfs路径,实验的输入路径是/workspace/dataguru/hadoopdev/week13/fpg/in/user2items.csv
-o 输出路径,指定hdfs中的输出路径
完整參数说明參见下表:
- 命令运行以后的输出文件夹:
casliyang@singlehadoop:~$ hadoop dfs -ls /workspace/dataguru/hadoopdev/week13/fpg/out
DEPRECATED: Use of this script to execute hdfs command is deprecated.
Instead use the hdfs command for it.
Found 4 items
-rw-r--r-- 3 casliyang supergroup 5567 2014-06-17 17:50 /workspace/dataguru/hadoopdev/week13/fpg/out/fList
drwxr-xr-x - casliyang supergroup 0 2014-06-17 17:51 /workspace/dataguru/hadoopdev/week13/fpg/out/fpgrowth
drwxr-xr-x - casliyang supergroup 0 2014-06-17 17:51 /workspace/dataguru/hadoopdev/week13/fpg/out/frequentpatterns
drwxr-xr-x - casliyang supergroup 0 2014-06-17 17:50 /workspace/dataguru/hadoopdev/week13/fpg/out/parallelcounting
挖掘出来的频繁模式在frequentpatterns目录下
casliyang@singlehadoop:~$ hadoop dfs -ls /workspace/dataguru/hadoopdev/week13/fpg/out/frequentpatterns
DEPRECATED: Use of this script to execute hdfs command is deprecated.
Instead use the hdfs command for it.
Found 2 items
-rw-r--r-- 3 casliyang supergroup 0 2014-06-17 17:51 /workspace/dataguru/hadoopdev/week13/fpg/out/frequentpatterns/_SUCCESS
-rw-r--r-- 3 casliyang supergroup 10017 2014-06-17 17:51 /workspace/dataguru/hadoopdev/week13/fpg/out/frequentpatterns/part-r-00000
该文件是序列化文件,不能直接查看,mahout提供了命令能够将其转换为普通文本:
mahout seqdumper -s /workspace/dataguru/hadoopdev/week13/fpg/out/frequentpatterns/part-r-00000 -o /home/casliyang/outpattern
这里要注意。-o指定的输出文件路径必须是linux文件系统。而且目标文件必须提前创建好,否则会报错。
- 终于输出到/home/casliyang/outpattern的部分结果
Key: 29099: Value: ([29099],18), ([29099, 4479],3)
Key: 29202: Value: ([29202],3)
Key: 29203: Value: ([29203],9), ([14020, 29203],3)
Key: 29224: Value: ([29224],3)
Key: 29547: Value: ([29547],5)
Key: 2963: Value: ([2963],8), ([2963, 21146],3)
Key: 2999: Value: ([2999],3)
Key: 3032: Value: ([3032],4)
Key: 3047: Value: ([3047],4)
Key: 3151: Value: ([3151],7), ([14020, 3151],4)
Key: 3181: Value: ([3181],3)
Key: 3228: Value: ([3228],14)
Key: 3313: Value: ([3313],3)
Key: 3324: Value: ([3324],3)
Key: 3438: Value: ([3438],3)
Key: 3458: Value: ([3458],4)
Key: 3627: Value: ([3627],11), ([3627, 11176],3)
。。。。
。
。
含义:
Key:itemid
Value:和该item相关的频繁模式及其支持度
有了挖掘出来的频繁模式。就能够进一步用程序依据业务需求做处理了。
Mahout真是个伟大的开源项目。