SQL题目记录

1.商品推荐题目

1.思路：

• 通过取差集 得出要推荐的商品
• 差集的选取：except直接取差集 或者a left join b on where b = null

2.知识点

1.except

select
  friendship_info.user1_id as user_id,
  sku_id
from
  friendship_info
  join favor_info on friendship_info.user2_id = favor_info.user_id
except
select
  user_id,
  sku_id
from
  favor_info

2.交并差

2.连续日期

1. 思路:

1.通过lag（向上） lead（向下）求出2天，三天比较

2.通过rk和日期进行相减

• 其实就是连续的日期进行分组,将连续的放入一组,不连续的放入另一组
• 第一步, 使用开窗 构造rk
• 计算日期和rk的差值,算出差值日期,这个差值日期就保证同一个分组,
• 使用开窗count 算出每个分组是否有 符合要求的 >=2

  with
  s1 as (
    select
      user_id,
      date (login_ts) start_date,
      row_number() OVER (
        PARTITION by
          user_id
        order by
          date (login_ts)
      ) rk
    from
      user_login_detail
    GROUP by
      user_id,
      date (login_ts)
  ),
  s11 as (
    select
      user_id,
      start_date,
      date_sub (start_date, rk) head_date
    from
      s1
  )
select
  user_id,
  min(start_date) `start_date`,
  max(start_date) `end_date`
from  s11 
GROUP by user_id,
  head_date
HAVING datediff(max(start_date) ,min(start_date))>=1

3. 截止到某天统计最近几天的总和和平均值

1.思路

其实想法都是一样的，要求截止到当天，最近3天的数据，就要把前两天的数据拿过来，
方法有两个，一个是自连接join 一个是直接开窗

1. join

join需要考虑空值，分母就少了，所以需要一个if判断

2.开窗函数

开窗直接sum avg 但是首先注意，这里时间是String类型，无法进行between and
需要转化成日期类型或者时间戳（秒 毫秒 都是int类型）
注意range的用法，range 的between and 是根据 前面的值进行减法 加法
比如 order by timestamp_day RANGE BETWEEN 172800 PRECEDING and current row 就是 范围是 当前行的timestamp_day -172800 到当前行 如果后面用following就是加

with
  s1 as (
    select
      create_date,
      sum(total_amount) total_Day,
      unix_timestamp(date(create_date)) timestamp_day
    from
      order_info
    GROUP by
      create_date
  )
select
	create_date,

    cast(sum(total_Day) over(order by timestamp_day RANGE BETWEEN  172800 PRECEDING and current row ) as decimal(16,2)) total_3d,
    cast(avg(total_Day) over(order by timestamp_day RANGE BETWEEN  172800 PRECEDING and current row ) as DECIMAl(16,2) )avg_3d
from s1

4. 关于奇数偶数的分配问题

1. 思路

1.直接将奇数和偶数分开然后union

with s1 as (select
	category_id,
    price,
    count(*) over(PARTITION by category_id) cnt ,
    row_number() over(partition by category_id order by price) rk
from sku_info)

-- 偶数
select
	category_id,
    cast(avg(if(rk=(cnt/2) or rk = (cnt/2)+1,price,null)) as DECIMAL(16,2)) medprice
from s1
where cnt%2=0
GROUP by category_id
union
-- 奇数 
select
	category_id,
    -- 奇数是向上取整取1个, 偶数是向上取整,然后再向上找一个
    cast(avg(if(rk=ceil(cnt/2),price,null)) as DECIMAL(16,2)) medprice
from s1
where cnt%2=1
GROUP by category_id

2.优化 因为奇数是取结果向下+1, 偶数是取结果和结果+1,可以过滤

因为rk是整数
奇数rk = floor(cnt/2)+1 cnt/2<floor(cnt/2)+1 <cnt/2+1
偶数rk = floor(cnt/2)+1 floor(cnt/2) 两边相等

如果把条件改 大于等于 小于等于即可

with s1 as (select
	category_id,
    price,
    count(*) over(PARTITION by category_id) cnt ,
    row_number() over(partition by category_id order by price) rk
from sku_info)

-- 偶数
select
	category_id,
    cast(avg(price) as DECIMAL(16,2)) medprice
from s1
where cnt/2<=rk  and rk<= (cnt/2+1)  
GROUP by category_id