Hive窗口函数/分析函数详解
小数志
共 17458字，需浏览 35分钟
·
2021-03-04 10:01
在sql中有一类函数叫做聚合函数,例如sum()、avg()、max()等等,这类函数可以将多行数据按照规则聚集为一行,一般来讲聚集后的行数是要少于聚集前的行数的。但是有时我们想要既显示聚集前的数据,又要显示聚集后的数据,这时我们便引入了窗口函数。窗口函数又叫OLAP函数/分析函数，窗口函数兼具分组和排序功能。
窗口函数最重要的关键字是 partition by 和 order by。
具体语法如下：over (partition by xxx order by xxx)
sum,avg,min,max 函数
准备数据
1建表语句:
2create table bigdata_t1(
3cookieid string,
4createtime string, --day
5pv int
6) row format delimited
7fields terminated by ',';
8
9加载数据：
10load data local inpath '/root/hivedata/bigdata_t1.dat' into table bigdata_t1;
11
12cookie1,2018-04-10,1
13cookie1,2018-04-11,5
14cookie1,2018-04-12,7
15cookie1,2018-04-13,3
16cookie1,2018-04-14,2
17cookie1,2018-04-15,4
18cookie1,2018-04-16,4
19
20开启智能本地模式
21SET hive.exec.mode.local.auto=true;
SUM函数和窗口函数的配合使用：结果和ORDER BY相关,默认为升序。
1#pv1
2select cookieid,createtime,pv,
3sum(pv) over(partition by cookieid order by createtime) as pv1
4from bigdata_t1;
5
6#pv2
7select cookieid,createtime,pv,
8sum(pv) over(partition by cookieid order by createtime rows between unbounded preceding and current row) as pv2
9from bigdata_t1;
10
11#pv3
12select cookieid,createtime,pv,
13sum(pv) over(partition by cookieid) as pv3
14from bigdata_t1;
15
16#pv4
17select cookieid,createtime,pv,
18sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and current row) as pv4
19from bigdata_t1;
20
21#pv5
22select cookieid,createtime,pv,
23sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and 1 following) as pv5
24from bigdata_t1;
25
26#pv6
27select cookieid,createtime,pv,
28sum(pv) over(partition by cookieid order by createtime rows between current row and unbounded following) as pv6
29from bigdata_t1;
30
31
32pv1: 分组内从起点到当前行的pv累积，如，11号的pv1=10号的pv+11号的pv, 12号=10号+11号+12号
33pv2: 同pv1
34pv3: 分组内(cookie1)所有的pv累加
35pv4: 分组内当前行+往前3行，如，11号=10号+11号， 12号=10号+11号+12号，
36 13号=10号+11号+12号+13号， 14号=11号+12号+13号+14号
37pv5: 分组内当前行+往前3行+往后1行，如，14号=11号+12号+13号+14号+15号=5+7+3+2+4=21
38pv6: 分组内当前行+往后所有行，如，13号=13号+14号+15号+16号=3+2+4+4=13，
39 14号=14号+15号+16号=2+4+4=10
如果不指定rows between,默认为从起点到当前行;
如果不指定order by，则将分组内所有值累加;
关键是理解rows between含义,也叫做window子句：
preceding：往前
following：往后
current row：当前行
unbounded：起点
unbounded preceding 表示从前面的起点
unbounded following：表示到后面的终点
AVG，MIN，MAX，和SUM用法一样。
row_number,rank,dense_rank,ntile函数
准备数据
1cookie1,2018-04-10,1
2cookie1,2018-04-11,5
3cookie1,2018-04-12,7
4cookie1,2018-04-13,3
5cookie1,2018-04-14,2
6cookie1,2018-04-15,4
7cookie1,2018-04-16,4
8cookie2,2018-04-10,2
9cookie2,2018-04-11,3
10cookie2,2018-04-12,5
11cookie2,2018-04-13,6
12cookie2,2018-04-14,3
13cookie2,2018-04-15,9
14cookie2,2018-04-16,7
15
16CREATE TABLE bigdata_t2 (
17cookieid string,
18createtime string, --day
19pv INT
20) ROW FORMAT DELIMITED
21FIELDS TERMINATED BY ','
22stored as textfile;
23
24加载数据：
25load data local inpath '/root/hivedata/bigdata_t2.dat' into table bigdata_t2;
ROW_NUMBER()使用
ROW_NUMBER()从1开始，按照顺序，生成分组内记录的序列。
1SELECT
2cookieid,
3createtime,
4pv,
5ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn
6FROM bigdata_t2;
RANK 和 DENSE_RANK使用
RANK() 生成数据项在分组中的排名，排名相等会在名次中留下空位 。
DENSE_RANK()生成数据项在分组中的排名，排名相等会在名次中不会留下空位。
1SELECT
2cookieid,
3createtime,
4pv,
5RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn1,
6DENSE_RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn2,
7ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv DESC) AS rn3
8FROM bigdata_t2
9WHERE cookieid = 'cookie1';
NTILE
有时会有这样的需求:如果数据排序后分为三部分，业务人员只关心其中的一部分，如何将这中间的三分之一数据拿出来呢?NTILE函数即可以满足。
ntile可以看成是：把有序的数据集合平均分配到指定的数量（num）个桶中, 将桶号分配给每一行。如果不能平均分配，则优先分配较小编号的桶，并且各个桶中能放的行数最多相差1。
然后可以根据桶号，选取前或后 n分之几的数据。数据会完整展示出来，只是给相应的数据打标签；具体要取几分之几的数据，需要再嵌套一层根据标签取出。
1SELECT
2cookieid,
3createtime,
4pv,
5NTILE(2) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn1,
6NTILE(3) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn2,
7NTILE(4) OVER(ORDER BY createtime) AS rn3
8FROM bigdata_t2
9ORDER BY cookieid,createtime;
其他一些窗口函数
lag,lead,first_value,last_value 函数
LAG
LAG(col,n,DEFAULT) 用于统计窗口内往上第n行值第一个参数为列名，第二个参数为往上第n行（可选，默认为1），第三个参数为默认值（当往上第n行为NULL时候，取默认值，如不指定，则为NULL）
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAG(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS last_1_time,
6 LAG(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS last_2_time
7 FROM bigdata_t4;
8
9
10 last_1_time: 指定了往上第1行的值，default为'1970-01-01 00:00:00'
11 cookie1第一行，往上1行为NULL,因此取默认值 1970-01-01 00:00:00
12 cookie1第三行，往上1行值为第二行值，2015-04-10 10:00:02
13 cookie1第六行，往上1行值为第五行值，2015-04-10 10:50:01
14 last_2_time: 指定了往上第2行的值，为指定默认值
15 cookie1第一行，往上2行为NULL
16 cookie1第二行，往上2行为NULL
17 cookie1第四行，往上2行为第二行值，2015-04-10 10:00:02
18 cookie1第七行，往上2行为第五行值，2015-04-10 10:50:01
LEAD
与LAG相反
LEAD(col,n,DEFAULT) 用于统计窗口内往下第n行值
第一个参数为列名，第二个参数为往下第n行（可选，默认为1），第三个参数为默认值（当往下第n行为NULL时候，取默认值，如不指定，则为NULL）
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LEAD(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS next_1_time,
6 LEAD(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS next_2_time
7 FROM bigdata_t4;
FIRST_VALUE
取分组内排序后，截止到当前行，第一个值
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS first1
6 FROM bigdata_t4;
LAST_VALUE
取分组内排序后，截止到当前行，最后一个值
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1
6 FROM bigdata_t4;
如果想要取分组内排序后最后一个值，则需要变通一下：
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1,
6 FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime DESC) AS last2
7 FROM bigdata_t4
8 ORDER BY cookieid,createtime;
特别注意order by
如果不指定ORDER BY，则进行排序混乱，会出现错误的结果
1 SELECT cookieid,
2 createtime,
3 url,
4 FIRST_VALUE(url) OVER(PARTITION BY cookieid) AS first2
5 FROM bigdata_t4;
cume_dist,percent_rank 函数
这两个序列分析函数不是很常用，注意：序列函数不支持WINDOW子句
数据准备
1 d1,user1,1000
2 d1,user2,2000
3 d1,user3,3000
4 d2,user4,4000
5 d2,user5,5000
6
7 CREATE EXTERNAL TABLE bigdata_t3 (
8 dept STRING,
9 userid string,
10 sal INT
11 ) ROW FORMAT DELIMITED
12 FIELDS TERMINATED BY ','
13 stored as textfile;
14
15 加载数据：
16 load data local inpath '/root/hivedata/bigdata_t3.dat' into table bigdata_t3;
CUME_DIST 和order by的排序顺序有关系
CUME_DIST 小于等于当前值的行数/分组内总行数 order 默认顺序 正序 升序
比如，统计小于等于当前薪水的人数，所占总人数的比例
1 SELECT
2 dept,
3 userid,
4 sal,
5 CUME_DIST() OVER(ORDER BY sal) AS rn1,
6 CUME_DIST() OVER(PARTITION BY dept ORDER BY sal) AS rn2
7 FROM bigdata_t3;
8
9 rn1: 没有partition,所有数据均为1组，总行数为5，
10 第一行：小于等于1000的行数为1，因此，1/5=0.2
11 第三行：小于等于3000的行数为3，因此，3/5=0.6
12 rn2: 按照部门分组，dpet=d1的行数为3,
13 第二行：小于等于2000的行数为2，因此，2/3=0.6666666666666666
PERCENT_RANK
PERCENT_RANK 分组内当前行的RANK值-1/分组内总行数-1
1 SELECT
2 dept,
3 userid,
4 sal,
5 PERCENT_RANK() OVER(ORDER BY sal) AS rn1, --分组内
6 RANK() OVER(ORDER BY sal) AS rn11, --分组内RANK值
7 SUM(1) OVER(PARTITION BY NULL) AS rn12, --分组内总行数
8 PERCENT_RANK() OVER(PARTITION BY dept ORDER BY sal) AS rn2
9 FROM bigdata_t3;
10
11 rn1: rn1 = (rn11-1) / (rn12-1)
12 第一行,(1-1)/(5-1)=0/4=0
13 第二行,(2-1)/(5-1)=1/4=0.25
14 第四行,(4-1)/(5-1)=3/4=0.75
15 rn2: 按照dept分组，
16 dept=d1的总行数为3
17 第一行，(1-1)/(3-1)=0
18 第三行，(3-1)/(3-1)=1
grouping sets,grouping__id,cube,rollup 函数
这几个分析函数通常用于OLAP中，不能累加，而且需要根据不同维度上钻和下钻的指标统计，比如，分小时、天、月的UV数。
数据准备
1 2018-03,2018-03-10,cookie1
2 2018-03,2018-03-10,cookie5
3 2018-03,2018-03-12,cookie7
4 2018-04,2018-04-12,cookie3
5 2018-04,2018-04-13,cookie2
6 2018-04,2018-04-13,cookie4
7 2018-04,2018-04-16,cookie4
8 2018-03,2018-03-10,cookie2
9 2018-03,2018-03-10,cookie3
10 2018-04,2018-04-12,cookie5
11 2018-04,2018-04-13,cookie6
12 2018-04,2018-04-15,cookie3
13 2018-04,2018-04-15,cookie2
14 2018-04,2018-04-16,cookie1
15
16 CREATE TABLE bigdata_t5 (
17 month STRING,
18 day STRING,
19 cookieid STRING
20 ) ROW FORMAT DELIMITED
21 FIELDS TERMINATED BY ','
22 stored as textfile;
23
24 加载数据：
25 load data local inpath '/root/hivedata/bigdata_t5.dat' into table bigdata_t5;
GROUPING SETS
grouping sets是一种将多个group by 逻辑写在一个sql语句中的便利写法。
等价于将不同维度的GROUP BY结果集进行UNION ALL。
GROUPING__ID，表示结果属于哪一个分组集合。
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 GROUPING SETS (month,day)
9 ORDER BY GROUPING__ID;
10
11 grouping_id表示这一组结果属于哪个分组集合，
12 根据grouping sets中的分组条件month，day，1是代表month，2是代表day
13
14 等价于
15 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month UNION ALL
16 SELECT NULL as month,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day;
再如：
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 GROUPING SETS (month,day,(month,day))
9 ORDER BY GROUPING__ID;
10
11 等价于
12 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month
13 UNION ALL
14 SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day
15 UNION ALL
16 SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM bigdata_t5 GROUP BY month,day;
CUBE
根据GROUP BY的维度的所有组合进行聚合。
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 WITH CUBE
9 ORDER BY GROUPING__ID;
10
11 等价于
12 SELECT NULL,NULL,COUNT(DISTINCT cookieid) AS uv,0 AS GROUPING__ID FROM bigdata_t5
13 UNION ALL
14 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month
15 UNION ALL
16 SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day
17 UNION ALL
18 SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM bigdata_t5 GROUP BY month,day;
ROLLUP
是CUBE的子集，以最左侧的维度为主，从该维度进行层级聚合。
1 比如，以month维度进行层级聚合：
2 SELECT
3 month,
4 day,
5 COUNT(DISTINCT cookieid) AS uv,
6 GROUPING__ID
7 FROM bigdata_t5
8 GROUP BY month,day
9 WITH ROLLUP
10 ORDER BY GROUPING__ID;
11
12 --把month和day调换顺序，则以day维度进行层级聚合：
13
14 SELECT
15 day,
16 month,
17 COUNT(DISTINCT cookieid) AS uv,
18 GROUPING__ID
19 FROM bigdata_t5
20 GROUP BY day,month
21 WITH ROLLUP
22 ORDER BY GROUPING__ID;
23 （这里，根据天和月进行聚合，和根据天聚合结果一样，因为有父子关系，如果是其他维度组合的话，就会不一样）
评论