hadoop学习-stream-Top K记录

求海量数据中最大的K个记录

来源于《hadoop实战》(hadoop in action)（美 拉姆）第4.7章节有关stream的习题。

数据源：apat63_99.txt 专利描述数据集，包含专利号、专利申请年份等等信息。可从美国国家经济研究局获得，网址为http://www.nber.org/patents 

大约有290万条记录。

这里的脚本用的是python。

apat63_99.txt里面存有专利的各种信息，这里以第9列的专利特定专利声明个数作为排序的key值，将最大的K条完整记录输出。

apat63_99.txt格式：

[plain] view plaincopy

1. "PATENT","GYEAR","GDATE","APPYEAR","COUNTRY","POSTATE","ASSIGNEE","ASSCODE","CLAIMS","NCLASS","CAT","SUBCAT","CMADE","CRECEIVE","RATIOCIT","GENERAL","ORIGINAL","FWDAPLAG","BCKGTLAG","SELFCTUB","SELFCTLB","SECDUPBD","SECDLWBD"  
2. 3070801,1963,1096,,"BE","",,1,,269,6,69,,1,,0,,,,,,,  
3. 3070802,1963,1096,,"US","TX",,1,,2,6,63,,0,,,,,,,,,  
4. 3070803,1963,1096,,"US","IL",,1,,2,6,63,,9,,0.3704,,,,,,,  
5. 3070804,1963,1096,,"US","OH",,1,,2,6,63,,3,,0.6667,,,,,,,  
6. 3070805,1963,1096,,"US","CA",,1,,2,6,63,,1,,0,,,,,,,  
7. 3070806,1963,1096,,"US","PA",,1,,2,6,63,,0,,,,,,,,,  
8. 3070807,1963,1096,,"US","OH",,1,,623,3,39,,3,,0.4444,,,,,,,  
9. 3070808,1963,1096,,"US","IA",,1,,623,3,39,,4,,0.375,,,,,,,  
10. 3070809,1963,1096,,"US","AZ",,1,,4,6,65,,0,,,,,,,,,  

Top_K.py:

[python] view plaincopy

1. #!/usr/bin/env python  
2. #encoding:utf-8  
3.   
4. import sys  
5.   
6. # Top_K   
7. Top_K = int(sys.argv[1])  
8.   
9. #源数据apat63_99.txt的第9列，表示特定专利声明个数  
10. index=8  
11.   
12. #初始化top_k数组,用来保存记录  
13. top=[]  
14. topresult=[]  
15. for i in range(Top_K):  
16.     top.append(0)  
17.     topresult.append("")  
18.   
19. #插入记录  
20. def add(temp,record):  
21.     global index  
22.     global Top_K  
23.     global top  
24.     global topresult  
25.     i=0  
26.     if temp>top[0]:  
27.         top[0]=temp  
28.         topresult[0]=record  
29.         for i in range(Top_K-1):  
30.             if temp>top[i+1]:  
31.                 top[i]=top[i+1]  
32.                 topresult[i]=topresult[i+1]  
33.             else:  
34.                 i=i-1  
35.                 break  
36.   
37.         i=i+1  
38.         top[i]=temp  
39.         topresult[i]=record  
40.   
41. for line in sys.stdin:  
42.     fields = line.strip().split(",")  
43.     if fields[index].isdigit():  
44.         val = int(fields[index])  
45.         add(val,line.strip())  
46.   
47. #输出top_k记录  
48. i=0  
49. for i in range(Top_K-1,-1,-1):  
50.     print topresult[i]  

对每条读入的记录，通过add函数判断并添加到topresult链表中。

add函数第一个参数temp，是用来排序的key值，这里传进去的也就是数据源中的第9列的值，即index=8

第二个参数是key值对应的完整专利描述记录。

在map阶段和reduce阶段都调用此脚本。即可完成Top K记录的查找。

运行指令：

[plain] view plaincopy

1. bin/hadoop jar contrib/streaming/hadoop-streaming-1.2.1.jar -input Top_K/apat63_99.txt -output output -mapper 'Top_K.py 8' -file Top_K.py -reducer 'Top_K.py 8' -file Top_K.py  

参数中的8 表示希望求得最大的8条记录。

运行结果：

4533693,1985,9349,1983,"US","CA",538840,2,346,524,1,15,13,80,1,0.6444,0.6272,7.8875,6.7692,0.0909,0.0769,0.0127,0.0125
5812453,1998,14144,1994,"JP","",581270,3,348,365,2,24,22,0,1,,0.1694,,13.1818,0.1818,0.1818,,
4068298,1978,6584,1975,"US","OH",557440,2,393,707,2,22,12,24,1,0.7292,0.8194,12.7083,5.6667,0,0,0,0
4624109,1986,9825,1983,"US","CA",,1,394,60,4,45,3,2,1,0.5,0.4444,9,8.6667,,,,
4081478,1978,6661,1976,"US","MI",600480,2,472,564,1,14,7,1,0.8571,0,0.8333,4,9.5714,0.6,0.4286,1,1
4373527,1983,8446,1979,"US","MD",295920,2,642,604,3,32,6,49,1,0.414,0.4444,10.6939,6.8333,0,0,0.1087,0.102
4085139,1978,6682,1976,"US","MI",600480,2,706,564,1,14,7,1,0.8571,0,0.8333,14,9.5714,0.6,0.4286,0,0
5095054,1992,11757,1990,"DE","",618615,2,868,524,1,15,40,111,0.975,0.8442,0.8586,4.1622,14.45,0.0571,0.05,0.0093,0.009

参考资料：

http://blog.csdn.net/jtlyuan/article/details/7560905

《hadoop实战》(hadoop in action)（美 拉姆）