目录
一、窗口函数的知识点
1.1 窗户函数的定义
1.2 窗户函数的语法
1.3 窗口函数分类
聚合函数
排序函数
前后函数
头尾函数
1.4 聚合函数
二、实际案例
2.1 每个用户累积访问次数
0 问题描述
1 数据准备
2 数据分析
3 小结
2.2 各直播间最大的同时在线人数
0 问题描述
1 数据准备
2 数据分析
3 小结
2.3 历史至今每个小时内同时在线人数
0 问题描述
1 数据准备
2 数据分析
3 小结
2.4 某个时间段、每个小时内同时在线人数
0 问题描述
1 数据准备
2 数据分析
3 小结
2.5 学生各学科的成绩
0 问题描述
1 数据准备
2 数据分析
3 小结
一、窗口函数的知识点
1.1 窗户函数的定义
窗口函数可以拆分为【窗口+函数】。窗口函数官网指路:
LanguageManual WindowingAndAnalytics - Apache Hive - Apache Software Foundationhttps://cwiki.apache.org/confluence/display/Hive/LanguageManual%20WindowingAndAnalytics
- 窗口:限定函数的计算范围(窗口函数:针对分组后的数据,从逻辑角度指定计算的范围,并没有从物理上真正的切分,只有group by 是物理分组,真正意义上的分组)
- 函数:计算逻辑
- 窗口函数的位置:跟sql里面聚合函数的位置一样,from -> join -> on -> where -> group by->select 后面的普通字段,窗口函数 -> having -> order by -> lmit 。 窗口函数不能跟聚合函数同时出现。聚合函数包括count、sum、 min、max、avg。
- sql 执行顺序:from -> join -> on -> where -> group by->select 后面的普通字段,聚合函数-> having -> order by -> limit
1.2 窗户函数的语法
<窗口函数>window_name over ( [partition by 字段...] [order by 字段...] [窗口子句] )
- window_name:给窗口指定一个别名。
- over:用来指定函数执行的窗口范围,如果后面括号中什么都不写,即over() ,意味着窗口包含满足where 条件的所有行,窗口函数基于所有行进行计算。
- 符号[] 代表:可选项; | : 代表二选一
- partition by 子句: 窗口按照哪些字段进行分组,窗口函数在不同的分组上分别执行。分组间互相独立。
- order by 子句:每个partition内部按照哪些字段进行排序,如果没有partition ,那就直接按照最大的窗口排序,且默认是按照升序(asc)排列。
- 窗口子句:显示声明范围(不写窗口子句的话,会有默认值)。常用的窗口子句如下:
rows between unbounded preceding and unbounded following; -- 上无边界到下无边界(一般用于求 总和)rows between unbounded preceding and current row; --上无边界到当前记录(累计值)rows between 1 preceding and current row; --从上一行到当前行rows between 1 preceding and 1 following; --从上一行到下一行rows between current row and 1 following; --从当前行到下一行
ps: over()里面有order by子句,但没有窗口子句时 ,即: <窗口函数> over ( partition by 字段... order by 字段... ),此时窗口子句是有默认值的 --> rows between unbounded preceding and current row (上无边界到当前行)。
此时窗口函数语法:<窗口函数> over ( partition by 字段... order by 字段... ) 等价于
<窗口函数> over ( partition by 字段... order by 字段... rows between unbounded preceding and current row)
需要注意有个特殊情况:当order by 后面跟的某个字段是有重复行的时候, <窗口函数> over ( partition by 字段... order by 字段... ) 不写窗口子句的情况下,窗口子句的默认值是:range between unbounded preceding and current row(上无边界到当前相同行的最后一行)。
因此,遇到order by 后面跟的某个字段出现重复行,且需要计算【上无边界到当前行】,那就需要手动指定窗口子句 rows between unbounded preceding and current row ,偷懒省略窗口子句会出问题~
ps: 窗口函数的执行顺序是在where之后,所以如果where子句需要用窗口函数作为条件,需要多一层查询,在子查询外面进行。
【例如】求出登录记录出现间断的用户Id
selectid
from (selectid,login_date,lead(login_date, 1, '9999-12-31')over (partition by id order by login_date) next_login_date--窗口函数 lead(向后取n行)--lead(column1,n,default)over(partition by column2 order by column3) 查询当前行的后边第n行数据,如果没有就为nullfrom (--用户在同一天可能登录多次,需要去重selectid,date_format(`date`, 'yyyy-MM-dd') as login_datefrom user_loggroup by id, date_format(`date`, 'yyyy-MM-dd')) tmp1) tmp2
where datediff(next_login_date, login_date) >=2
group by id;
1.3 窗口函数分类
哪些函数可以是窗口函数呢?(放在over关键字前面的)
sum(column) over (partition by .. order by .. 窗口子句);
count(column) over (partition by .. order by .. 窗口子句);
max(column) over (partition by .. order by .. 窗口子句);
min(column) over (partition by .. order by .. 窗口子句);
avg(column) over (partition by .. order by .. 窗口子句);
ps : 高级聚合函数:
collect_list 收集并形成list集合,结果不去重;
collect_set 收集并形成set集合,结果去重;
举例:
--每个月的入职人数以及姓名select
month(replace(hiredate,'/','-')),count(*) as cnt,collect_list(name) as name_list
from employee
group by month(replace(hiredate,'/','-'));/*
输出结果
month cn name_list
4 2 ["宋青书","周芷若"]
6 1 ["黄蓉"]
7 1 ["郭靖"]
8 2 ["张无忌","杨过"]
9 2 ["赵敏","小龙女"]
*/
-
排序函数
-- 顺序排序——1、2、3
row_number() over(partition by .. order by .. )-- 并列排序,跳过重复序号——1、1、3(横向加)
rank() over(partition by .. order by .. )-- 并列排序,不跳过重复序号——1、1、2(纵向加)
dense_rank() over(partition by .. order by .. )
-
前后函数
-- 取得column列的前n行,如果存在则返回,如果不存在,返回默认值default
lag(column,n,default) over(partition by order by) as lag_test
-- 取得column列的后n行,如果存在则返回,如果不存在,返回默认值default
lead(column,n,default) over(partition by order by) as lead_test
-
头尾函数
---当前窗口column列的第一个数值,如果有null值,则跳过
first_value(column,true) over (partition by ..order by.. 窗口子句) ---当前窗口column列的第一个数值,如果有null值,不跳过
first_value(column,false) over (partition by ..order by.. 窗口子句)--- 当前窗口column列的最后一个数值,如果有null值,则跳过
last_value(column,true) over (partition by ..order by.. 窗口子句) --- 当前窗口column列的最后一个数值,如果有null值,不跳过
last_value(column,false) over (partition by ..order by.. 窗口子句)
1.4 聚合函数
sum() /count() /max() /min() /avg() 函数,一般用于开窗求累积汇总值。
sum(column) over (partition by .. order by .. 窗口子句);
count(column) over (partition by .. order by .. 窗口子句);
max(column) over (partition by .. order by .. 窗口子句);
min(column) over (partition by .. order by .. 窗口子句);
avg(column) over (partition by .. order by .. 窗口子句);
二、实际案例
2.1 每个用户累积访问次数
0 问题描述
统计每个用户累积访问次数
1 数据准备
create table if not exists table6
(userid string comment '用户id',visitdate string comment '访问时间',visitcount int comment '访问次数'
)comment '用户访问次数';
2 数据分析
selectuserid,visit_date,vc1,--再求出用户历史至今的累积访问次数sum(vc1) over (partition by userid order by visit_date ) as vc2
from ( --先求出用户每个月的累积访问次数selectuserid,date_format(visitdate, 'yyyy-MM') as visit_date,sum(visitcount) as vc1from table6group by userid, date_format(visitdate, 'yyyy-MM')) tmp1;
3 小结
2.2 各直播间最大的同时在线人数
0 问题描述
根据直播间的用户访问记录,统计各直播间最大的同时在线人数。
1 数据准备
create table if not exists table7
(room_id int comment '直播间id',user_id int comment '用户id',login_time string comment '用户进入直播间时间',logout_time string comment '用户离开直播间时间'
)comment '直播间的用户访问记录';
INSERT overwrite table table7
VALUES (1,100,'2021-12-01 19:00:00', '2021-12-01 19:28:00'),(1,100,'2021-12-01 19:30:00', '2021-12-01 19:53:00'),(2,100,'2021-12-01 21:01:00', '2021-12-01 22:00:00'),(1,101,'2021-12-01 19:05:00', '2021-12-01 20:55:00'),(2,101,'2021-12-01 21:05:00', '2021-12-01 21:58:00'),(1,102,'2021-12-01 19:10:00', '2021-12-01 19:25:00'),(2,102,'2021-12-01 19:55:00', '2021-12-01 21:00:00'),(3,102,'2021-12-01 21:05:00', '2021-12-01 22:05:00'),(1,104,'2021-12-01 19:00:00', '2021-12-01 20:59:00'),(2,104,'2021-12-01 21:57:00', '2021-12-01 22:56:00'),(2,105,'2021-12-01 19:10:00', '2021-12-01 19:18:00'),(3,106,'2021-12-01 19:01:00', '2021-12-01 21:10:00');
2 数据分析
selectroom_id,max(num)
from (selectroom_id,sum(flag) over (partition by room_id order by dt) as numfrom (selectroom_id,user_id,login_time as dt,--对登入该直播间的人,标记 11 as flagfrom table7unionselectroom_id,user_id,logout_time as dt,--对退出该直播间的人,标记 -1-1 as flagfrom table7) tmp1) tmp2
--求出直播间最大的同时在线人数
group by room_id;
3 小结
该题的关键点在于:对每个用户进入/退出直播间的行为进行打标签,再利用sum()over聚合函数计算最终的数值。
2.3 历史至今每个小时内同时在线人数
由案例2.2 引申出来的案例 2.3和 案例2.4
0 问题描述
根据直播间用户访问记录,不限制时间段,统计历史至今的各直播间每个小时内的同时在线人数
1 数据准备
create table if not exists table7
(room_id int comment '直播间id',user_id int comment '用户id',login_time string comment '用户进入直播间时间',logout_time string comment '用户离开直播间时间'
)comment '直播间的用户访问记录';
INSERT overwrite table table7
VALUES (1,100,'2021-12-01 19:00:00', '2021-12-01 19:28:00'),(1,100,'2021-12-01 19:30:00', '2021-12-01 19:53:00'),(2,100,'2021-12-01 21:01:00', '2021-12-01 22:00:00'),(1,101,'2021-12-01 19:05:00', '2021-12-01 20:55:00'),(2,101,'2021-12-01 21:05:00', '2021-12-01 21:58:00'),(1,102,'2021-12-01 19:10:00', '2021-12-01 19:25:00'),(2,102,'2021-12-01 19:55:00', '2021-12-01 21:00:00'),(3,102,'2021-12-01 21:05:00', '2021-12-01 22:05:00'),(1,104,'2021-12-01 19:00:00', '2021-12-01 20:59:00'),(2,104,'2021-12-01 21:57:00', '2021-12-01 22:56:00'),(2,105,'2021-12-01 19:10:00', '2021-12-01 19:18:00'),(3,106,'2021-12-01 19:01:00', '2021-12-01 21:10:00');
2 数据分析
完整代码如下:
with temp_data as (selectroom_id,user_id,login_time,logout_time,hour(login_time) as min_time,-- hour('2021-12-01 19:30:00') = 19hour(logout_time) as max_time,length(space(hour(logout_time) - hour(login_time))) as lg,split(space(hour(logout_time) - hour(login_time)), '') as disfrom table7
)selectroom_id,on_time,count(1) as cnt
from (select distinctroom_id,user_id,min_time,max_time,dis,dis_index,(min_time + dis_index) as on_timefrom temp_data lateral view posexplode(dis) n as dis_index,dis_dataorder by user_id,min_time,max_time,dis,dis_index) tmp1
group by room_id, on_time
order by room_id, on_time;
代码拆解分析:
--以一条数据为例,room_id user_id login_time logout_time1 100 '2021-12-01 19:00:00' '2021-12-01 21:28:00'
(1)上述数据取时间hour(login_time) as min_time 、hour(logout_time)as max_time1(room_id),100(user_id),19(min_time),21(max_time)
(2)split(space(hour(logout_time) - hour(login_time)), '') 的结果:根据[21-19]=2,利用space函数生成长度是2的空格字符串,再用split拆分1(room_id),100(user_id),19(min_time),21(max_time),['','','']
(3)用posexplode经过转换增加行(列转行,炸裂),通过下角标index来获取 on_time时间,根据数组['','',''],得到index的取值是0,1,2炸裂得出下面三行数据(一行变三行)1(room_id),100(user_id),19(min_time),19 = 19+0 (on_time = min_time+index)1(room_id),100(user_id),19(min_time),20 = 19+1 (on_time = min_time+index)1(room_id),100(user_id),19(min_time),21 = 19+2 (on_time = min_time+index)炸裂的目的:将用户在线的时间段[19-21] 拆分成具体的小时,19,20,21;
(4)根据room_id,on_time进行分组,求出每个直播间分时段的在线人数
3 小结
上述代码中用到的函数有:
一、空格字符串函数:space
语法:space(int n)
返回值:string
说明:返回值是n的空格字符串
select length (space(10)) --> 10二、split函数(分割字符串)
语法:split(string str,string pat)
返回值:array
说明:按照pat字符串分割str,会返回分割后的字符串数组
例如: select split ('abcdf','c') from test; -> ["ab","df"]三、炸裂函数explode 语法:lateral view explode(split(a,',')) tmp as new_column返回值:string说明:按照分隔符切割字符串,并将数组中内容炸裂成多行字符串举例:select student_score from test lateral view explode(split(student_score,',')) tmp as student_scoreposexplode语法:lateral view posexploed(split(a,',')) tmp as pos,item 返回值:string说明:按照分隔符切割字符串,并将数组中内容炸裂成多行字符串(炸裂具备瞎下角标 0,1,2,3)举例:select student_name, student_score from testlateral view posexplode(split(student_name,',')) tmp1 as student_name_index,student_namelateral view posexplode(split(student_score,',')) tmp2 as student_score_index,student_scorewhere student_score_index = student_name_index
2.4 某个时间段、每个小时内同时在线人数
0 问题描述
根据直播间用户访问记录,统计某个时间段的各直播间每个小时内的同时在线人数,假设时间段是['2021-12-01 19:00:00', '2021-12-01 23:00:00']
1 数据准备
create table if not exists table7
(room_id int comment '直播间id',user_id int comment '用户id',login_time string comment '用户进入直播间时间',logout_time string comment '用户离开直播间时间'
)comment '直播间的用户访问记录';
INSERT overwrite table table7
VALUES (1,100,'2021-12-01 19:00:00', '2021-12-01 19:28:00'),(1,100,'2021-12-01 19:30:00', '2021-12-01 19:53:00'),(2,100,'2021-12-01 21:01:00', '2021-12-01 22:00:00'),(1,101,'2021-12-01 19:05:00', '2021-12-01 20:55:00'),(2,101,'2021-12-01 21:05:00', '2021-12-01 21:58:00'),(1,102,'2021-12-01 19:10:00', '2021-12-01 19:25:00'),(2,102,'2021-12-01 19:55:00', '2021-12-01 21:00:00'),(3,102,'2021-12-01 21:05:00', '2021-12-01 22:05:00'),(1,104,'2021-12-01 19:00:00', '2021-12-01 20:59:00'),(2,104,'2021-12-01 21:57:00', '2021-12-01 22:56:00'),(2,105,'2021-12-01 19:10:00', '2021-12-01 19:18:00'),(3,106,'2021-12-01 19:01:00', '2021-12-01 21:10:00');
2 数据分析
完整代码如下:
with temp_data1 as (selectroom_id,user_id,login_time,logout_time,hour(login_time) as min_time,hour(logout_time) as max_time,split(space(hour(logout_time) - hour(login_time)), '') as disfrom table7where login_time >= '2021-12-01 19:00:00'and login_time <= '2021-12-01 21:00:00'
)selectroom_id,on_time,count(1) as cnt
from (select distinctroom_id,user_id,min_time,max_time,dis_index,(min_time + dis_index) as on_timefrom temp_data1 lateral view posexplode(dis) n1 as dis_index, dis_dataorder by user_id,min_time,max_time,dis_index) tmp
group by room_id, on_time
order by room_id, on_time;
3 小结
解题思路与2.3一致,只需要限制下时间区间
2.5 学生各学科的成绩
0 问题描述
基于不同的窗口限定范围,统计各学生的学科成绩。
1 数据准备
create table if not exists table9
(name string comment '学生名称',subject string comment '学科',score int comment '分数'
)comment '学生分数';
INSERT overwrite table table9
VALUES ('a','数学',12),('b','数学',19),('c','数学',17),('d','数学',24),('a','英语',77),('c','英语',11),('d','英语',34),('a','语文',61);
2 数据分析
selectname,subject,score,--1.全局聚合sum(score) over () as sum1,--2.根据学科分组,组内全局聚合sum(score) over (partition by subject) as sum2,--3.根据学科分组,根据分数排序,计算由起点到当前行的累积值sum(score) over (partition by subject order by score) as sum3,--4.根据学科分组,根据分数排序,计算由起点到当前行的累积值 (sum3跟sum4的结果是一样的)sum(score) over (partition by subject order by score rows between unbounded preceding and current row ) as sum4,--5.根据学科分组,根据分数排序,计算上一行到当前行的累积值sum(score) over (partition by subject order by score rows between 1 preceding and current row ) as sum5,--6.根据学科分组,根据分数排序,计算上一行到下一行的累积值sum(score) over (partition by subject order by score rows between 1 preceding and 1 following) as sum6,--7.根据学科分组,根据分数排序,计算当前行到后面所有行的累积值sum(score) over (partition by subject order by score rows between current row and unbounded following ) as sum7
from table9;
3 小结
窗口函数 = 窗口+ 函数,解题时需要梳理清楚函数的计算范围。