Hive SQL 中有很多窗口函数值得我们在平时的数据开发处理中好好使用。通常包含排序类、聚合类、累计计算,等。在数据开发的
此篇就简单罗列一些窗口函数的SQL例子,做一个复习回顾。
rank,dense_rank, row_number, ntile,等。
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
select
*
,rank() over(partition by `班级` order by `成绩` desc) as `班级排名`
,rank() over (order by `成绩` desc) as ranking -- 相同分数,相同排名,接下来的跳过+n排名(不连续排名)
,dense_rank() over (order by `成绩` desc) as dese_ranking -- 相同分数,相同排名,接下来的继续+1排名(连续排名)
,row_number() over (order by `成绩` desc) as row_num -- 按成绩倒序,编号(排序)
from database_name.table_a;
假设:按照学生学号 student_no(每个学生有唯一学号),统计每个学生的 总分、平均分、目前考试科目、最高分、最低分。
select
*
,sum(score) over (partition by student_no) as current_sum -- 每个学生总成绩
,avg(score) over (partition by student_no) as current_avg -- 平均分
,count(score) over (partition by student_no) as current_cnt
,max(score) over (partition by student_no) as current_max
,min(score) over (partition by student_no) as current_min
from database_name.table_a;
按照学号升序排序,对成绩score统计一些累计指标。
select
*
,sum(score) over (order by student_no) as current_sum -- 累计求和
,avg(score) over (order by student_no) as current_avg
,count(score) over (order by student_no) as current_cnt
,max(score) over (order by student_no) as current_max
,min(score) over (order by student_no) as current_min
from database_name.table_a;
按照 成绩 倒序排序,rank()是连续排名, 1, 2, 3, 4,。。。遇到相同分数,继续排序下去。
select
*
,rank() over(order by `成绩` desc) as ranking
from database_name.table_a;
对数据进行偏移计算
主要考虑关键词: rows n preceding, rows n following, rows between [n|unbounded] preceding and [n|current|unbounded] following。或者 lag(offset) over(partition by order by ),lead(offset) over(partition by order by )。
指定最靠近的3行作为汇总对象: rows n preceding:包含本条记录的前n行记录
select
*
,avg(sale_price) over(order by product_id rows 2 preceding) as moving_avg -- 移动平均
from database_name.table_a;
指定包含本条记录,前1行,后1行 记录的 sale_price平均值
select
*
,avg(sale_price) over(order by product_id rows 1 preceding and 1 following) as moving_avg -- 移动平均
from database_name.table_a;
SELECT
cookieid,
createtime,
pv,
SUM(pv) OVER(PARTITION BY cookieid ORDER BY createtime) AS pv1, -- 默认为从起点到当前行
SUM(pv) OVER(PARTITION BY cookieid ORDER BY createtime ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS pv2, -- 从起点到当前行,结果同pv1
SUM(pv) OVER(PARTITION BY cookieid) AS pv3, -- 分组内所有行
SUM(pv) OVER(PARTITION BY cookieid ORDER BY createtime ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) AS pv4, -- 当前行+往前3行
SUM(pv) OVER(PARTITION BY cookieid ORDER BY createtime ROWS BETWEEN 3 PRECEDING AND 1 FOLLOWING) AS pv5, -- 当前行+往前3行+往后1行
SUM(pv) OVER(PARTITION BY cookieid ORDER BY createtime ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS pv6 -- 当前行+往后所有行
FROM win1
order by createtime;
对数据的偏移统计,还可以用到
LAG(expr, offset, default_value) over([partition by col1] [order by col2] [desc|asc]) ,表示向前偏移offset行记录,按照over后面的条件,进行expr统计,如果偏移offset后越界,可以是null, 也可以是指定的default_value。
LEAD(expr, offset, default_value) over([partition by col1] [order by col2] [desc|asc]) ,表示向后偏移offset行记录,按照over后面的条件,进行expr统计,如果偏移offset后越界,可以是null, 也可以是指定的default_value。
LAG(字段,n,为空时的默认值)向上n行
SELECT
cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LAG(createtime, 1, '1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS last_1_time,
LAG(createtime, 2) OVER(PARTITION BY cookieid ORDER BY createtime) AS last_2_time
FROM win4;
LEAD(字段,n,为空时的默认值)向下n行
SELECT
cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
LEAD(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS next_1_time,
LEAD(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS next_2_time
FROM win4;
select
seq,
LAG(seq+100, 1, -1) over (partition by window order by seq) as r1
from sliding_window;
--
select
c_Double_a,c_String_b,c_int_a,
lead(c_int_a, 1) over(partition by c_Double_a order by c_String_b) as r2
from dual;
select
c_String_a,c_time_b,c_Double_a,
lead(c_Double_a,1) over(partition by c_String_a order by c_time_b) as r3
from dual;
select
c_String_in_fact_num,c_String_a,c_int_a,
lead(c_int_a) over(partition by c_String_in_fact_num order by c_String_a) as r4
from dual;
分片函数
NTILE分片函数,随机分配n个编号给相应的分组。
ntile 用于将分组数据按照顺序切分成n片,并返回当前切片值。如果切片不均匀,默认增加第一个切片的分布。
SELECT
cookieid,
createtime,
pv,
NTILE(2) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn1, -- 分组内将数据分成2片
NTILE(3) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn2, -- 分组内将数据分成3片
NTILE(4) OVER(ORDER BY createtime) AS rn3 -- 将所有数据分成4片
FROM win2
ORDER BY
cookieid,
createtime;
-- 现在需要将所有职工根据部门按sal高到低切分为3组,并获得职工自己组内的序号。
select
deptno
,ename
,sal
,ntile(3)over(partition by deptno order by sal desc) as nt3
from emp;
PERCENT_RANK 返回位置百分数,按照rank进行排序之后,找出当前位置的rank所在的位置百分数
SELECT
dept,
userid,
sal,
PERCENT_RANK() OVER(ORDER BY sal) AS rn1, -- 分组内
RANK() OVER(ORDER BY sal) AS rn11, -- 分组内RANK值
SUM(1) OVER(PARTITION BY NULL) AS rn12, -- 分组内总行数
PERCENT_RANK() OVER(PARTITION BY dept ORDER BY sal) AS rn2
FROM win3;
FIRST_VALUE()取分组后截止到当前行相应字段的第一个值
SELECT cookieid,
createtime,
url,
ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS first1,
LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime DESC) AS first2,
LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1
FROM lxw1234;
NTH_VALUE 用于计算第n个值。如果n超过窗口的最大长度,返回NULL。
select
user_id,
price,
nth_value(price, 2) over(partition by user_id) as nth_value
from test_src;
CLUSTER_SAMPLE 分组抽样
-- 从每组中抽取10%的样本
select
key, value
from (
select
key, value,
cluster_sample(10, 1) over(partition by key) as flag
from tbl
) sub
where flag=true;
CUME_DIST 累计分布
-- 现在需要将所有职工根据部门分组,再求sal在同一组内的前百分之几。
select
deptno
,ename
,sal
,concat(round(cume_dist(sal) over(partition by deptno order by sal desc)*100, 2), '%') as cume_dist
from emp;
此外,还有一些针对字符串的常用操作,具体可以参考阿里云Maxcompute文档。
WM_CONCAT 用指定的separator做分隔符,链接str中的值。
-- 对表进行分组排序后合并: 对表test按照id列进行分组排序,并将同组的内容进行合并。
SELECT
id,
WM_CONCAT('',alphabet) as res
FROM test
GROUP BY
id
ORDER BY
id
LIMIT 100;
-- collect_list(col) 在给定Group内,将col指定的表达式聚合为一个数组。
需要对应的配置
set project odps.sql.type.system.odps2=true;
或
set odps.sql.type.system.odps2=true;
COLLECT_SET(col) 在给定Group内,将col指定的表达式聚合为一个无重复元素的集合数组,输出类型是set(集合)。
以下为一些常用的统计方差函数。
-- variance/var_pop(col) 计算 col 列方差
-- var_samp 用于计算指定数字列的样本方差。
-- covar_pop 用于计算指定两个数字列的总体协方差。
-- COVAR_SAMP 用于计算指定两个数字列的样本协方差。
PERCENTILE 返回指定列精确的第p位百分数。p必须在0和1之间。
-- DOUBLE PERCENTILE(BIGINT col, p)
SELECT
PERCENTILE(c1,0),
PERCENTILE(c1,0.3),
PERCENTILE(c1,0.5),
PERCENTILE(c1,1)
FROM var_test;
以上就是复习常用的窗口函数和一些字符串统计函数。还有比较实用的是 行转列(一行转多行)、列转行等数据处理,可以参考explode、lateral view 关键词,及 explode 与 split、lateral view 等 并用。
以下图为 表 table_explode_lateral_view 的记录。
需求1: 只拆解 goods_id 字段
select
explode(split(goods_id, ',')) as new_goods_id
from table_explode_lateral_view;
需求2: 只拆解area字段
select
explode(split(area, ',')) as new_area
from table_explode_lateral_view;
LATERAL VIEW的使用:
侧视图的意义是配合explode(或者其他的UDTF),一个语句生成把单行数据拆解成多行后的数据结果集。
select
goods_id2,
sale_info
from table_explode_lateral_view
LATERAL VIEW explode(split(goods_id,','))goods as goods_id2;
其中LATERAL VIEW explode(split(goods_id,',')) goods相当于一个虚拟表,与原表explode_lateral_view笛卡尔积关联。
也可以多重使用
select
goods_id2,
sale_info,
area2
from table_explode_lateral_view
LATERAL VIEW explode(split(goods_id,',')) goods as goods_id2
LATERAL VIEW explode(split(area,',')) area as area2;
也是三个表笛卡尔积的结果。
需求3: 解析以上sale_info成二维表。
select
get_json_object(concat('{',sale_info_1,'}'),'$.source') as source,
get_json_object(concat('{',sale_info_1,'}'),'$.monthSales') as monthSales,
get_json_object(concat('{',sale_info_1,'}'),'$.userCount') as monthSales,
get_json_object(concat('{',sale_info_1,'}'),'$.score') as monthSales
from table_explode_lateral_view
LATERAL VIEW explode(split(regexp_replace(regexp_replace(sale_info,'\\[\\{',''),'}]',''),'},\\{')) sale_info as sale_info_1;
也可以使用json_tuple对一个json字段/或者数组,同时对多个key进行解析。
select
sale_info_1
from table_explode_lateral_view
lateral view json_tuple(sale_info,'source','monthSales','userCount', 'score') sale_info_new as sale_info_1;
更多细节,还需要根据实际情况进行处理。