一、聚合函数
对一组值进行计算,并返回单个值,也被称为组函数;
-
聚合计算过程:选定分组字段 – 分组 – 计算:
1.选定分组字段:在 select 里;
2.分组:用 group by;
3.计算:根据实际情况,选定聚合函数; -
聚合函数
1.计数:count()
2.求和:sum()
3.最大值/最小值:max/min()
4.平均值:avg() -
聚合值过滤
聚合值过滤:having,不是where! -
目前SQL执行顺序:from – where – group by – having – select
-
eg:按省份和高考分数分段,统计高分段的人数:
安徽学生,高考分数>620;
江苏学生,高考分数>610;
其他省份学生,高考分数>600;
二、举例练习
表例--常规聚合函数使用:
--注意事项:hive中distinct和group by 不能一起用
--1、查询每个学生考试单科最高分是多少?字段:学号、姓名、单科最高分;
select
stu_id as 学号
,name as 姓名
,max(score) as 单科最高分
from score_info
group by stu_id,name;
--2、查询每个学生考试单科最低分是多少?字段:学号、姓名、单科最低分;
select
stu_id as 学号
,name as 姓名
,min(score) as 单科最低分
from score_info
group by stu_id,name;
--3、查询每个学生考试平均分是多少?字段:学号、姓名、考试平均分;
select
stu_id as 学号
,name as 姓名
,avg(score) as 考试平均分
from score_info
group by stu_id,name;
--4、统计每个学生考了几科?字段:学号、姓名、考试科数;
select
stu_id as 学号
,name as 姓名
,count(distinct subject_id) as 考试科数
from score_info
group by stu_id,name;
--5、统计每个学生考试总分是多少?字段:学号、姓名、考试总分;
select
stu_id as 学号
,name as 姓名
,sum(score) as 考试总分
from score_info
group by stu_id,name;
--6、查询每个省份,不同学院的人数有多少?字段:省份、学院、人数;
select
from_where as 省份
,college as 学院
,count(distinct stu_id) as 人数
from student_info
group by from_where,college
order by from_where,college;
--7、查询每个班级人数有多少?字段:学院、专业、班级、人数;
select
college as 学院
,major as 专业
,class as 班级
,count(stu_id) as 人数
from class_info
group by college,major,class
order by college,major;
--与其他函数结合使用:
--1、查询高数=100分 和 物理=100分的人数有多少?字段:高数满分人数,物理满分人数;
--if
select
count(if(subject='高数' and score=100,stu_id,null)) as 高数满分人数
,count(if(subject='物理' and score=100,stu_id,null)) as 物理满分人数
from score_info;
--case when
count(distinct (case when subject='高数' and score=100 then stu_id else null end)) as 高数满分人数
,count(distinct (case when subject='物理' and score=100 then stu_id else null end)) as 物理满分人数
from score_info;
--2、查询每个班级,高数和物理不及格的人数有多少(<60分)?字段:学院、专业、班级、高数不及格人数、物理不及格人数;
select
college as 学院
,major as 专业
,class as 班级
,count(if(subject='高数' and score<60,subject,null)) as 高数不及格人数
,count(if(subject='物理' and score<60,subject,null)) as 物理不及格人数
from score_info join class_info
on score_info.stu_id=class_info.stu_id
group by college,major,class
order by college,major;
--
select
subject
,case when score>0 and score<60 then '0-60'
when score>=60 and score <70 then '60-70'
when score>=70 and score <80 then '70-80'
when score>=80 and score <80 then '80-90'
when score>=90 and score <100 then '90-100'
end as diff_part
,count(distinct stu_id) as 人数
from score_info
group by subject,diff_part;
--hive中,group by不能用别名(diff_part),需要全部复制过来
网友评论