SQL Server：在组间（而不是组内）使用Lead/Lag分析函数

Question

非常抱歉我的帖子很长，但是我已经提供了复制并粘贴的样本数据和一种可能的解决方法。问题相关部分在帖子的上半部分（横线之上）。

我有如下表格：

 Dt          customer_id  buy_time     money_spent
 -------------------------------------------------
 2000-01-04  100          11:00:00.00  2
 2000-01-05  100          16:00:00.00  1
 2000-01-10  100          13:00:00.00  4
 2000-01-10  100          14:00:00.00  3
 2000-01-04  200          09:00:00.00  10
 2000-01-06  200          10:00:00.00  11
 2000-01-06  200          11:00:00.00  5
 2000-01-10  200          08:00:00.00  20

需要一条查询语句来获取这个结果集

 Dt          Dt_next     customer_id  buy_time     money_spent
 -------------------------------------------------------------
 2000-01-04  2000-01-05  100          11:00:00.00  2
 2000-01-05  2000-01-10  100          16:00:00.00  1
 2000-01-10  NULL        100          13:00:00.00  4
 2000-01-10  NULL        100          14:00:00.00  3
 2000-01-04  2000-01-06  200          09:00:00.00  10
 2000-01-06  2000-01-10  200          10:00:00.00  11
 2000-01-06  2000-01-10  200          11:00:00.00  5
 2000-01-10  NULL        200          08:00:00.00  20

那是：我希望对于每个客户(customer_id)和每一天(Dt)，在下一个日期该客户访问的时间(Dt_next)。

我已经有一个查询可以得到后面的结果集(数据和查询包含在下方的水平线下)。但是，它涉及到一个left outer join和两个dense_rank聚合函数。这种方法对我来说似乎有点笨拙，而我认为应该有更好的解决方案。非常感谢任何提供替代方案的指针! 谢谢!

顺便说一下: 我正在使用SQL Server 11，表格中有>>1m条目。

---

我的查询:

 select
   customer_table.Dt
   ,customer_table_lead.Dt as Dt_next
   ,customer_table.customer_id
   ,customer_table.buy_time
   ,customer_table.money_spent
 from
 (
   select 
     #customer_data.*
     ,dense_rank() over (partition by customer_id order by customer_id asc, Dt asc) as Dt_int
   from #customer_data
 ) as customer_table
 left outer join
 (
   select distinct
     #customer_data.Dt
     ,#customer_data.customer_id
     ,dense_rank() over (partition by customer_id order by customer_id asc, Dt asc)-1 as Dt_int
   from #customer_data
 ) as customer_table_lead
 on
 (
   customer_table.Dt_int=customer_table_lead.Dt_int
   and customer_table.customer_id=customer_table_lead.customer_id
 )

示例数据：

 create table #customer_data (
   Dt date not null,
   customer_id int not null,
   buy_time time(2) not null,
   money_spent float not null
 );

 insert into #customer_data values ('2000-01-04',100,'11:00:00',2);
 insert into #customer_data values ('2000-01-05',100,'16:00:00',1);
 insert into #customer_data values ('2000-01-10',100,'13:00:00',4);
 insert into #customer_data values ('2000-01-10',100,'14:00:00',3);

 insert into #customer_data values ('2000-01-04',200,'09:00:00',10);
 insert into #customer_data values ('2000-01-06',200,'10:00:00',11);
 insert into #customer_data values ('2000-01-06',200,'11:00:00',5);
 insert into #customer_data values ('2000-01-10',200,'08:00:00',20);

- cryo111

Answer 1

尝试使用以下查询：

select cd.Dt
    , t.Dt_next
    , cd.customer_id
    , cd.buy_time
    , cd.money_spent
from (
    select Dt
        , LEAD(Dt) OVER (PARTITION BY customer_id ORDER BY Dt) AS Dt_next
        , customer_id
    from (
        select distinct Dt, customer_id
        from #customer_data
    ) t
) t
inner join #customer_data cd on t.customer_id = cd.customer_id and t.Dt = cd.Dt

为什么字段money_spent是浮点数类型？这可能会导致计算问题。请将其转换为十进制类型。

- GriGrim

---

刚刚检查了一下，程序可以正常运行，而且查询语句也更加简短！我还会在更大的表格上测试性能。感谢您提供浮点类型的提示（这只是我在创建示例数据时首先想到的类型 - 实际表格与此处提供的示例数据不同）。 - cryo111

在寻找类似需求的帖子时，我发现了这篇文章，而在我的情况下，PARTITION BY 是答案。也谢谢你 :) - Onur Omer