Question

美好的一天，

我在SSMS中使用MS T-SQL。我试图确定一种基于变量或case语句的窗口函数的方法。我需要窗口继续循环，直到某个结果发生，然后通过这些观察切割窗口。另外，也许有更好的方法在Windows之外做这个，我已经研究了WHILE表达式作为潜在的选项。

以下是一个示例数据集：

   CREATE TABLE #Items ([Transaction] int, [Item] varchar(10), [Price] money, [Price2] varchar(25), [SaleDate] smalldatetime)

   INSERT INTO   #Items ([Transaction], [Item], [Price], [Price2], [SaleDate])
   VALUES (123,'Blue',1100,'Blue1100','04/10/2018'),
   (124,'Blue',1100,'Blue1100','04/09/2018'),
   (125,'Blue',1100,'Blue1100','04/09/2018'),
   (126,'Blue',3000,'Blue3000','03/27/2018'),
   (127,'Blue',1100,'Blue1100','03/27/2018'),
   (128,'Blue',1100,'Blue1100','03/27/2018'),
   (129,'Blue',1100,'Blue1100','03/27/2018'),
   (130,'Blue',1100,'Blue1100','03/27/2018'),
   (131,'Red',3328,'Red3328','04/12/2018'),
   (132,'Red',3328,'Red3328','04/09/2018'),
   (133,'Red',3328,'Red3328','04/06/2018'),
   (134,'Red',3328,'Red3328','04/04/2018'),
   (135,'Red',3328,'Red3328','04/02/2018'),
   (136,'Red',3328,'Red3328','04/02/2018'),
   (137,'Yellow',1340,'Yellow1340','04/09/2018'),
   (138,'Yellow',1340,'Yellow1340','04/08/2018'),
   (139,'Yellow',1340,'Yellow1340','04/08/2018'),
   (140,'Yellow',1500,'Yellow1500','04/05/2018'),
   (141,'Yellow',1340,'Yellow1340','04/05/2018'),
   (142,'Yellow',1340,'Yellow1340','04/05/2018'),
   (143,'Yellow',1340,'Yellow1340','04/02/2018'),
   (144,'Yellow',1340,'Yellow1340','03/31/2018')

在这种情况下，我希望窗口查看按[SaleDate] desc排序的每个[Item]，并按顺序创建具有相同[Price]的顺序的窗口大小。只要[价格]没有中断，窗口就会基于这些观察。一旦确定了，我想添加一个列，在窗口中为这些观察结果提供最小的[SaleDate]。

所需结果位于此表的最后一栏

我正在尝试使用以下查询，但无法弄清楚如何让窗口循环：

SELECT 
Transaction
, Item
, Price
,Price2
, SaleDate
,(CASE WHEN PRICE2 = LEAD(PRICE2) OVER (ORDER BY Price ASC, SaleDate DESC ) 
AND LEAD(PRICE2) OVER (ORDER BY Price ASC, SaleDate DESC ) = LEAD(PRICE2) OVER (ORDER BY Price ASC, SaleDate DESC ) 
THEN MIN(SaleDate) ELSE NULL END) 'StartDate'

FROM #Items

GROUP BY 
Transaction
, Item
, Price
,Price2
, SaleDate

欢迎任何和所有建议！谢谢！

Answer 1

这可以通过递归CTE完成。

示例SQL：

;with CTE as (
    SELECT [Transaction], Item, Price, Price2, SaleDate, 
    row_number() over (partition by Item order by SaleDate asc, [Transaction] desc) as rn
    FROM #Items
)
, RCTE as (
     SELECT 
       c.*, 
       SaleDate as SaleDatePriceChange
     from CTE c
     where rn = 1

     union all

     SELECT 
       c.*,
       case when c.Price = r.Price then r.SaleDatePriceChange else c.SaleDate end
     from RCTE r 
     join CTE c on c.Item = r.Item and c.rn = r.rn + 1   

)
select [Transaction], Item, Price, Price2, SaleDate, SaleDatePriceChange
from RCTE
order by Item, rn desc;

第一个CTE用于添加row_number。

这样可以更轻松地加入递归CTE中的下一条记录。

Answer 2

这是获取这些开始日期的另一种方法这次没有递归。

它使用LAG获取前一记录的价格这样以前的价格就可以与记录的价格进行比较如果价格发生变化，IIF将返回1，否则返回0。

SUM窗口函数将向上添加1和0，从而产生排名诀窍是，当添加0时，等级数不会增加。

然后FIRST_VALUE用于获取其中排名第一的SaleDate。

select 
  [Transaction], Item, Price, Price2, SaleDate, 
  first_value(t.SaleDate) over (partition by t.Item, q2.PriceChangeRank order by t.SaleDate, t.[Transaction] desc) as PriceChangedDate
from
(
    select TransactionID,
      sum(PriceChanged) over (partition by Item order by SaleDate, TransactionID desc) as PriceChangeRank
    from
    (
        select [Transaction] as TransactionID, Item, SaleDate,
          iif(Price != lag(Price) over (partition by Item order by SaleDate, [Transaction] desc),1,0)  as PriceChanged
        from #Items
    ) q1
) q2
join #Items t on (t.[Transaction] = q2.TransactionID)
order by [Transaction];

不确定此方法是否比使用动态SQL更快。但它是另一种选择。

窗口循环功能

2 个答案: