使用LAG函数获取不一定在上一行中的值

时间:2018-03-09 18:27:50

标签: sql-server sql-server-2012 lag

我对LAG的功能范围有疑问。在我正在处理的具体示例中,我有一个类似于以下的表,尽管这是非常简化的:

ID          Name             ItemID      VisitDate
----------- ---------------- ----------- ----------
316         Name,Test        9           2016-11-01
316         Name,Test        12          2016-11-01
316         Name,Test        89          2018-03-09
316         Name,Test        10          2018-03-09
316         Name,Test        1           2018-03-09

我的目标是使用LAG有一个额外的列,显示当前之前的最近的VisitDate。所以我希望看到的是

ID          Name             ItemID      VisitDate  LagDate
----------- ---------------- ----------- ---------- ----------
316         Name,Test        9           2016-11-01 NULL
316         Name,Test        12          2016-11-01 NULL
316         Name,Test        89          2018-03-09 2016-11-01
316         Name,Test        10          2018-03-09 2016-11-01
316         Name,Test        1           2018-03-09 2016-11-01

但是,我还没有办法让表格以这种方式出现。最初我写了以下代码

SELECT
    TT.ID
   ,TT.Name
   ,TT.ItemID
   ,TT.VisitDate
   ,LAG(TT.VisitDate) OVER ( PARTITION BY TT.ID ORDER BY TT.VisitDate ) AS LagDate
FROM
    @TestTable AS TT;

然而,这返回了下表:

ID          Name             ItemID      VisitDate  LagDate
----------- ---------------- ----------- ---------- ----------
316         Name,Test        9           2016-11-01 NULL
316         Name,Test        12          2016-11-01 2016-11-01
316         Name,Test        89          2018-03-09 2016-11-01
316         Name,Test        10          2018-03-09 2018-03-09
316         Name,Test        1           2018-03-09 2018-03-09

那时我意识到单个日期的多个ItemID值对我造成了问题。那么有没有人知道我仍然可以在这张桌子上使用LAG的方法,但是我想在每一行上找到合适的LagDate吗?

4 个答案:

答案 0 :(得分:1)

我看不到使用窗口函数解决这个问题的简单方法。我能想到的最简单的事情就是使用OUTER APPLY

SELECT  TT.ID, TT.Name, TT.ItemID, TT.VisitDate, X.VisitDate
FROM mytable AS TT  
OUTER APPLY (
  SELECT TOP 1 VisitDate
  FROM mytable AS T
  WHERE T.ID = TT.ID AND T.VisitDate < TT.VisitDate
  ORDER BY T.VisitDate DESC) AS X

Demo here

答案 1 :(得分:0)

我使用密集排名:

declare @t table (id int, _name varchar(1000),ItemID int, visitDte date)

insert into @t
values
(316,         'Name,Test',        9           ,'2016-11-01')
,(316,         'Name,Test',        12          ,'2016-11-01')
,(316,         'Name,Test' ,      89          ,'2018-03-09')
,(316,         'Name,Test',        10          ,'2018-03-09')
,(316,         'Name,Test',        1           ,'2018-03-09')

;with cte as
(
select *, dr= dense_rank() over (partition by ID order by visitDte) from @t
)
, CleanUp as (
select distinct ID, visitDte, dr
from cte)

select t.*
    ,Cleanup.visitDte PriorVisit
from cte t
left join CleanUp on t.id=CleanUp.id and t.dr-1=Cleanup.dr

结果:

id  _name   ItemID  visitDte    dr  PriorVisit
316 Name,Test   9   2016-11-01  1   NULL
316 Name,Test   12  2016-11-01  1   NULL
316 Name,Test   89  2018-03-09  2   2016-11-01
316 Name,Test   10  2018-03-09  2   2016-11-01
316 Name,Test   1   2018-03-09  2   2016-11-01

答案 2 :(得分:0)

我认为这样做会

declare @t table(ID int, Name varchar(10), ItemID int,  VisitDate date);
insert into @t values 
       (316,         'Name,Test',        9,          '2016-11-01')
     , (316,         'Name,Test',       12,          '2016-11-01')
     , (316,         'Name,Test',       89,          '2018-03-09')
     , (316,         'Name,Test',       10,          '2018-03-09')
     , (316,         'Name,Test',        1,          '2018-03-09');
select *  
     , (select max(VisitDate) from @t td where td.VisitDate < t.VisitDate) as dd
from @t t 
order by t.VisitDate;

答案 3 :(得分:0)

LEAD / LAG功能中还有另一个参数,它是偏移量。如果不使用,则默认值为1.

LEAD ( scalar_expression [ ,offset ] , [ default ] )   
OVER ( [ partition_by_clause ] order_by_clause )

在你的情况下,你需要找到“返回”行数的偏移量,你可以通过使用中间窗口计算来实现。

        ;WITH T (ID, Name, ItemID,VisitDate) AS
        (
        SELECT 316, 'Name,Test', 9, CAST('2016-11-01' AS DATE)
        UNION ALL
        SELECT 316, 'Name,Test', 12, CAST('2016-11-01' AS DATE)
        UNION ALL
        SELECT 316, 'Name,Test', 89, CAST('2018-03-09' AS DATE)
        UNION ALL
        SELECT 316, 'Name,Test', 10, CAST('2018-03-09' AS DATE)
        UNION ALL
        SELECT 316, 'Name,Test', 1, CAST('2018-03-09' AS DATE)
        )
    , [Intermediate] as 
       (
        SELECT * , COUNT(Id) OVER (PARTITION BY Id , VisitDate ORDER BY Id) LagVal
        FROM T
       ) 
    ,[Intermediate1] as
       (
        SELECT * , ISNULL(LAG(LagVal) OVER (PARTITION BY Id order by Id), LagVal) LagValFixed
        FROM [Intermediate]
       ) 
        SELECT * , LAG(VisitDate, LagValFixed) OVER (PARTITION BY Id ORDER BY VisitDate) PriorVisit
        FROM [Intermediate1]

正如您所看到的,LagValFixed在最后一个语句中被用作LAG函数接受的偏移值的参数。