hive sql 行列转换开窗函数炸裂函数

准备原始数据集

学生表 student.csv
讲师表 teacher.csv
课程表 course.csv
分数表 score.csv
员工表 emp.csv
雇员表 employee.csv
电影表 movie.txt

学生表 student.csv

001,彭于晏,1995-05-16,男
002,胡歌,1994-03-20,男
003,周杰伦,1995-04-30,男
004,刘德华,1998-08-28,男
005,唐国强,1993-09-10,男
006,陈道明,1992-11-12,男
007,陈坤,1999-04-09,男
008,吴京,1994-02-06,男
009,郭德纲,1992-12-05,男
010,于谦,1998-08-23,男
011,潘长江,1995-05-27,男
012,杨紫,1996-12-21,女
013,蒋欣,1997-11-08,女
014,赵丽颖,1990-01-09,女
015,刘亦菲,1993-01-14,女
016,周冬雨,1990-06-18,女
017,范冰冰,1992-07-04,女
018,李冰冰,1993-09-24,女
019,邓紫棋,1994-08-31,女
020,宋丹丹,1991-03-01,女

讲师表 teacher.csv

1001,张高数
1002,李体音
1003,王子文
1004,刘丽英

课程表 course.csv

01,语文,1003
02,数学,1001
03,英语,1004
04,体育,1002
05,音乐,1002

分数表 score.csv

员工表 emp.csv

7369,张三,研发,800.00,30
7499,李四,财务,1600.00,20
7521,王五,行政,1250.00,10
7566,赵六,销售,2975.00,40
7654,侯七,研发,1250.00,30
7698,马八,研发,2850.00,30
7782,金九,行政,2450.0,30
7788,银十,行政,3000.00,10
7839,小芳,销售,5000.00,40
7844,小明,销售,1500.00,40
7876,小李,行政,1100.00,10
7900,小元,讲师,950.00,30
7902,小海,行政,3000.00,10
7934,小红明,讲师,1300.00,30
7934,小红,讲师,1300.00,

雇员表 employee.csv

张无忌,男,1980/02/12,2022/08/09,销售,3000,12000,阿朱_小昭,张小无:8_张小忌:9
赵敏,女,1982/05/18,2022/09/10,行政,9000,2000,阿三_阿四,赵小敏:8
宋青书,男,1981/03/15,2022/04/09,研发,18000,1000,王五_赵六,宋小青:7_宋小书:5
周芷若,女,1981/03/17,2022/04/10,研发,18000,1000,王五_赵六,宋小青:7_宋小书:5
郭靖,男,1985/03/11,2022/07/19,销售,2000,13000,南帝_北丐,郭芙,5_郭襄:4
黄蓉,女,1982/12/13,2022/06/11,行政,12000,null,东邪_西毒,郭芙,5_郭襄:4
杨过,男,1988/01/30,2022/08/13,前台,5000,null,郭靖_黄蓉,杨小过:2
小龙女,女,1985/02/12,2022/09/24,前台,6000,null,张三_李四,杨小过:2

电影表 movie.txt

《疑犯追踪》-悬疑,动作,科幻,剧情
《Lie to me》-悬疑,警匪,动作,心理,剧情
《战狼2》-战争,动作,灾难

订单表 order.csv

1,1001,小元,2022-01-01,10
2,1002,小海,2022-01-02,15
3,1001,小元,2022-02-03,23
4,1002,小海,2022-01-04,29
5,1001,小元,2022-01-05,46
6,1001,小元,2022-04-06,42
7,1002,小海,2022-01-07,50
8,1001,小元,2022-01-08,50
9,1003,小辉,2022-04-08,62
10,1003,小辉,2022-04-09,62
11,1004,小猛,2022-05-10,12
12,1003,小辉,2022-04-11,75
13,1004,小猛,2022-06-12,80
14,1003,小辉,2022-04-13,94

创建数据库和数据表

create database chap06;
use chap06;

-- 学生表 student.csv
create external table student (
	stu_id string comment '学生ID',
	stu_name string comment '学生姓名',
    birthday string comment '出生日期',
    gender string comment '学生性别'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz03/student';

load data local inpath '/root/data/data02/student.csv' overwrite into table student;

select * from student;

-- 讲师表 teacher.csv
create external table teacher (
	tea_id string comment '课程ID',
	tea_name string comment '课程名称'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz03/teacher';

load data local inpath '/root/data/data02/teacher.csv' overwrite into table teacher;

select * from teacher;

-- 课程表 course.csv
create external table course (
	course_id string comment '课程ID',
	course_name string comment '课程名称',
    tea_id string comment '讲师ID'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz03/course';

load data local inpath '/root/data/data02/course.csv' overwrite into table course;

select * from course;

-- 分数表 score.csv
create external table score (
	stu_id string comment '学生ID',
	course_id string comment '课程ID',
    score int comment '成绩'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz03/score';

load data local inpath '/root/data/data02/score.csv' overwrite into table score;
select * from score;

-- 员工表 emp.csv
create external table emp (
	emp_id int comment '员工ID',
	emp_name string comment '员工姓名',
    emp_job string comment '员工岗位',
    emp_salary decimal(8,2) comment '员工薪资',
    dept_id int comment '员工隶属部门ID'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz01/emp';
load data local inpath '/root/data/data02/emp.csv' overwrite into table emp;
select * from emp;

-- 雇员表 employee.csv
create external table employee(
    name string comment '姓名',
    sex  string comment '性别',
    birthday string comment '出生年月',
    hiredate string comment '入职日期',
    job string comment '岗位',
    salary int comment '薪资',
    bonus int comment '奖金',
    friends array<string> comment '朋友',
    children map<string,int> comment '孩子'
)
 row format delimited fields terminated by ','
 collection items terminated by '_'
 map keys terminated by ':'
 lines terminated by '\n'
 stored as textfile
 location '/quiz04/employee';
load data local inpath '/root/data/data02/employee.csv' into table employee;
select * from employee;

-- 电影表 movie.txt
create external table movie(
    name string comment '电影名称',
    category string comment '电影分类'
)
 row format delimited fields terminated by '-'
 lines terminated by '\n'
 stored as textfile
 location '/quiz04/movie';
load data local inpath '/root/data/data02/movie.txt' into table movie;
select * from movie;

-- 订单表 order.csv
create external table `order`
(
    order_id     string comment '订单id',
    user_id      string comment '用户id',
    user_name    string comment '用户姓名',
    order_date   string comment '下单日期',
    order_amount int comment '订单金额'
)
 row format delimited fields terminated by ','
 lines terminated by '\n'
 stored as textfile
 location '/quiz04/order';
load data local inpath '/root/data/data02/order.csv' into table `order`;
select * from `order`;

行列转换

列转行

create table test (
    stu_name string,
    course_name string,
    score int
);
insert into test values ('张三','语文','80'),('张三','数学','90'), ('李四','语文','85'),('李四','数学','95');
select * from test;

select stu_name,
       max(case when course_name = '语文' then score end) as yuwen,
       max(case when course_name = '数学' then score end) as shuxue
       from test group by stu_name;

行列转行列转行

-- 多个值转为集合 collect_list 不会去重
select collect_list(emp_job) job_list from emp;
-- 多个值转为集合 collect_set 会去重
select collect_set(emp_job) job_set from emp;
-- size 获取结合中元素的数量
select size(collect_set(emp_job)) job_count from emp;
-- concat_ws 将多个数据 以分隔符形式 拼接 concat_ws(分隔符,数据1,数据2,...)
select concat_ws('-',collect_set(emp_job)) job_string from emp;
-- split 字符串切分 以分隔符切分字符串 为集合
select split(concat_ws('-',collect_set(emp_job)),'-') job_item from emp;

行专列

create table sales (
    emp_name string,
    january int,
    february int,
    march int
);
insert into sales values ('张三',1000,2000,3000),('李四',1500,2500,3500);
select * from sales;

行列转换行转列

将转换后的结果还原

select t1.emp_name,
    sale_list[0] january,
    sale_list[1] february,
    sale_list[2] march
    from(
        select t.emp_name,collect_list(sale) sale_list from(
            select emp_name,'january' yue, january sale from sales
            union all
            select emp_name,'february' yue,february sale from sales
            union all
            select emp_name,'march' yue,march sale from sales) t
            group by t.emp_name) t1;

UDF UDTF UDAF

UDF，即用户定义函数(user-defined function)，作用于单行数据，并且产生一个数据行作为输出。
Hive中大多数函数都属于这一类，比如数学函数和字符串函数。UDF函数的输入与输出值是1:1关系。

UDTF，即用户定义表生成函数（user-defined table-generating function），
作用于单行数据，并且产生多个数据行。UDTF函数的输入与输出值是1:n的关系。

UDAF，用户定义聚集函数（user-defined aggregate function），作用于多行数据，产生一个输出数据行。
Hive中像COUNT、MAX、MIN和SUM这样的函数就是聚集函数。UDAF函数的输入与输出值是n:1的关系。

explode

array

select explode(array('java','python','scala','go')) as course;

map

select explode(map('name','李昊哲','gender','1')) as (key,value);

posexplode

select posexplode(array('java','python','scala','go')) as (pos,course);

inline

select inline(array(named_struct('id',1,'name','李昊哲','gender','1'),
                named_struct('id',2,'name','李哲','gender','0'),
                named_struct('id',3,'name','李大宝','gender','1')))
    as (id,name,gender);

lateral view

select * from employee lateral view explode(friends) t as friend;

select * from employee lateral view explode(children) t as children_name,children_age;

select * from employee
    lateral view explode(friends) t1 as friend
    lateral view explode(children) t2 as children_name,children_age;

select name, sex, birthday, hiredate, job, salary, bonus, friend,children_name,children_age  from employee e
    lateral view explode(friends) t1 as friend
    lateral view explode(children) t2 as children_name,children_age;

UDTF 案例

根据电影信息表，统计各分类的电影数量

select cate,count(name) as quantity  from movie
    lateral view explode(split(category,',')) tmp as cate
    group by cate;

窗口函数(开窗函数)

能为每行数据划分一个窗口，然后对窗口范围内的数据进行计算，最后将计算结果返回给该行
Function(arg1,…, argn) OVER ([PARTITION BY <…>] [ORDER BY <…>] [<window_expression>])
其中Function(arg1,…, argn) 可以是下面分类中的任意一个
聚合函数：比如sum max min avg count等
分析函数：比如lead lag first_value last_value等
排序函数：比如row_number rank dense_rank等
OVER [PARTITION BY <…>] 类似于group by 用于指定分组每个分组你可以把它叫做窗口
如果没有PARTITION BY 那么整张表的所有行就是一组
[ORDER BY <…>] 用于指定每个分组内的数据排序规则支持ASC、DESC
[<window_expression>] 用于指定每个窗口中操作的数据范围默认是窗口中所有行

聚合函数

聚合函数
rows 基于行
range 基于值
函数() over(rows between and 3)

unbounded preceding 表示从前面的起点
number preceding 往前
current row 当前行
number following 往后
unbounded following 表示到后面的终点

统计每个用户截至每次下单的累计下单总额

select *,
       sum(order_amount) over (
           partition by user_id ,substr(order_date,1,7)
           order by order_date
           rows between unbounded preceding and current row
           ) sum_order_amount
       from `order`;

select *,
       sum(order_amount) over (
           partition by user_id ,substr(order_date,1,7)
           order by order_date
           rows unbounded preceding
           ) sum_order_amount
       from `order`;

统计每个用户截至每次下单的当月累积下单总额

select *,
       sum(order_amount) over (
           partition by user_id ,substr(order_date,1,7)
           order by order_date
           rows between unbounded preceding and unbounded following
           ) sum_order_amount
       from `order`;

最近三笔订单总金额

当前订单金额与前两笔订单金额的总和
当前订单金额与后两笔订单金额的总和
当前订单金额与前一笔订单和后一笔订单金额的总和

当前订单金额与前两笔订单金额的总和

select *,
       sum(order_amount) over (
           partition by user_id
           order by order_date
           rows 2 preceding
           ) sum_order_amount
       from `order`;

当前订单金额与后两笔订单金额的总和

select *,
       sum(order_amount) over (
           partition by user_id
           order by order_date
           rows 2 following
           ) sum_order_amount
       from `order`;

当前订单金额与前一笔订单和后一笔订单金额的总和

select *,
       sum(order_amount) over (
           partition by user_id
           order by order_date
           rows between 1 preceding and 1 following
           ) sum_order_amount
       from `order`;

分析函数 lag lead first_value last_value

lag lead

lag() over() 与 lead() over() 函数是跟偏移量相关的两个分析函数，
通过这两个函数可以在一次查询中取出同一字段的前 N 行的数据 (lag) 和后 N 行的数据 (lead) 作为独立的列,
从而更方便地进行进行数据过滤。这种操作可以代替表的自联接，并且 LAG 和 LEAD 有更高的效率。
over() 表示 lag() 与 lead() 操作的数据都在 over() 的范围内，可以使用 partition by 语句（用于分组） order by 语句（用于排序）。
partition by a order by b 表示以 a 字段进行分组，再以 b 字段进行排序，对数据进行查询。
例如：lag(field, num, defaultvalue) field 需要查找的字段，num 往前查找的 num 行的数据，defaultvalue 没有符合条件的默认值
例如：lead(field, num, defaultvalue) field 需要查找的字段，num 往后查找的 num 行的数据，defaultvalue 没有符合条件的默认值

统计每个用户每次下单距离上次下单相隔的天数（首次下单按0天算）

select order_id, user_id, user_name, order_date, order_amount from (
    select order_id, user_id, user_name, order_date, order_amount,
       lag(order_date,1,order_date) over (partition by user_id order by order_date) pre_order_date
       from `order`) t where datediff(order_date,pre_order_date) = 0;

每个用户每个月首笔订单时间

select order_id, user_id, user_name, order_date, order_amount from (
    select order_id, user_id, user_name, order_date, order_amount,
       lag(order_date,1,order_date) over (partition by user_id,substr(order_date,1,7) order by order_date) pre_order_date
       from `order`) t where datediff(order_date,pre_order_date) = 0;

每个用户每个月最后笔订单时间

select order_id, user_id, user_name, order_date, order_amount from (
    select order_id, user_id, user_name, order_date, order_amount,
       lead(order_date,1,order_date) over (partition by user_id,substr(order_date,1,7) order by order_date) next_order_date
       from `order`) t where datediff(order_date,next_order_date) = 0;

每个岗位先先入职的远哥和后入在的员工工资差

select name, sex, birthday, hiredate, job, salary, bonus, friends, children, new_salary,(salary - new_salary) salary_diff from (
    select name, sex, birthday, hiredate, job, salary, bonus, friends, children,
       lead(salary,1,salary) over (partition by job order by hiredate) new_salary
       from employee) t;

first_value last_value

first_value 取每个分区内某列的第一个值
语法：first_value(col,true/false) over (partition by col1 order by col2)
第二个参数为true，跳过空值（默认为false）

last_value 取每个分区内某列的最后一个值
语法：last_value(col,true/false) over (partition by col1 order by col2)
第二个参数为true，跳过空值（默认为false）

每个用户每个月首笔订单时间

select order_id, user_id, user_name, order_date, order_amount,
       first_value(order_date) over (partition by user_id,substr(order_date,1,7) order by order_date) first_order_value
       from `order`;

每个用户每个月最后笔订单时间

select order_id, user_id, user_name, order_date, order_amount,
       last_value(order_date) over (
           partition by user_id,substr(order_date,1,7) order by order_date
           rows between current row and unbounded following) last_order_value
       from `order`;

每个用户每个月首笔订单时间和最后笔订单时间

select order_id, user_id, user_name, order_date, order_amount,
        first_value(order_date) over (partition by user_id,substr(order_date,1,7) order by order_date) first_order_value,
        last_value(order_date) over (
           partition by user_id,substr(order_date,1,7) order by order_date
           rows between current row and unbounded following) last_order_value
       from `order`;

select order_id, user_id, user_name, order_date, order_amount, first_order_value, last_order_value from
    (select order_id, user_id, user_name, order_date, order_amount,
            first_value(order_date) over (partition by user_id,substr(order_date,1,7) order by order_date) first_order_value,
            last_value(order_date) over (
               partition by user_id,substr(order_date,1,7) order by order_date
               rows between current row and unbounded following) last_order_value
           from `order`) t where order_date = first_order_value or order_date = last_order_value;

排序函数

分组排序取TopN

查询各科成绩前五名的学生

select a.course_id,a.stu_id,a.score from score a
    left join score b
    on a.course_id = b.course_id and a.score <= b.score
    group by a.stu_id,a.course_id,a.score
    having count(a.stu_id) <=5
    order by a.course_id,a.score desc;

select S1.course_id,s1.stu_id,s1.score from score s1 where
    (select count(*) from score s2
        where s2.course_id=s1.course_id AND s2.score > s1.score
        ) <= 5 order by s1.course_id,s1.score desc;

row_number

row_number() over () 连续序号
over()里头的分组以及排序的执行晚于 where 、group by、order by 的执行。

select * from
    (select course_id, stu_id,  score,
        row_number() over (partition by course_id order by score desc ) as mum
        from score) t where mum <= 5;

rank

rank() over () 排名跳跃排序序号不是连续的

select * from
    (select course_id, stu_id,  score,
        rank() over (partition by course_id order by score desc ) as mum
        from score) t where mum <= 5;

dense_rank

dense_rank() over () 排名连续排序

select * from
    (select course_id, stu_id,  score,
        dense_rank() over (partition by course_id order by score desc ) as mum
        from score) t where mum <= 5;

每个月每个消费总金额前三名的用户

select order_id, user_id, user_name, order_date, order_amount, total_order_amount, rank_total_order_amount from
(select order_id, user_id, user_name, order_date, order_amount, total_order_amount,
    dense_rank() over (partition by substr(order_date,1,7) order by total_order_amount desc) rank_total_order_amount
    from (
        select order_id, user_id, user_name, order_date, order_amount,
           sum(order_amount) over(partition by substr(order_date,1,7),user_id order by order_date
               rows between unbounded preceding and unbounded following) total_order_amount
           from `order`) t) t1 where rank_total_order_amount <= 3;