Hive分组取Top N

Hive在0.11.0版本开始加入了row_number、rank、dense_rank分析函数,可以查询分组排序后的top值

 
说明:
row_number() over ([partition col1] [order by col2])
rank() over ([partition col1] [order by col2])
dense_rank() over ([partition col1] [order by col2])
它们都是根据col1字段分组,然后对col2字段进行排序,对排序后的每行生成一个行号,这个行号从1开始递增
col1、col2都可以是多个字段,用','分隔
 
区别:
1)row_number:不管col2字段的值是否相等,行号一直递增,比如:有两条记录的值相等,但一个是第一,一个是第二
2)rank:上下两条记录的col2相等时,记录的行号是一样的,但下一个col2值的行号递增N(N是重复的次数),比如:有两条并列第一,下一个是第三,没有第二
3)dense_rank:上下两条记录的col2相等时,下一个col2值的行号递增1,比如:有两条并列第一,下一个是第二
 
row_number可以实现分页查询
 
实例:
 
  1. hive> create table t(name string, sub string, score int) row format delimited fields terminated by '\t';
 
数据在附件的a.txt里
  1. a chinese 98
  2. a english 90
  3. d chinese 88
  4. c english 82
  5. c math 98
  6. b math 89
  7. b chinese 79
  8. z english 90
  9. z math 89
  10. z chinese 80
  11. e math 99
  12. e english 87
  13. d english 90
 
 
1、row_number
  1. hive (test)> select *, row_number() over (partition by sub order by score) as od from t; 
 
2、rank
  1. hive (test)> select *, rank() over (partition by sub order by score) as od from t; 
 
3、dense_ran
  1. hive (test)> select *, dense_rank() over (partition by sub order by score desc) from t;
 
业务实例:
统计每个学科的前三名
  1. select * from (select *, row_number() over (partition by sub order by score desc) as od from t ) t where od<=3;
语文成绩是80分的排名是多少
  1. hive (test)> select od from (select *, row_number() over (partition by sub order by score desc) as od from t ) t where sub='chinese' and score=80;
分页查询
  1. hive (test)> select * from (select *, row_number() over () as rn from t) t1 where rn between 1 and 5;
 
 
 
 
 
 





附件列表

 

posted @ 2015-06-08 14:27  lishouguang  阅读(11023)  评论(1编辑  收藏  举报