Question

使用以下数据

create table #ph (product int, [date] date, price int)
insert into #ph select 1, '20120101', 1
insert into #ph select 1, '20120102', 1
insert into #ph select 1, '20120103', 1
insert into #ph select 1, '20120104', 1
insert into #ph select 1, '20120105', 2
insert into #ph select 1, '20120106', 2
insert into #ph select 1, '20120107', 2
insert into #ph select 1, '20120108', 2
insert into #ph select 1, '20120109', 1
insert into #ph select 1, '20120110', 1
insert into #ph select 1, '20120111', 1
insert into #ph select 1, '20120112', 1

我想产生以下输出：

product | date_from | date_to  | price
  1     | 20120101  | 20120105 |   1
  1     | 20120105  | 20120109 |   2
  1     | 20120109  | 20120112 |   1

如果我按价格分组并显示最大和最小日期，那么我将得到以下不是我想要的（参见日期的重叠）。

product | date_from | date_to  | price
  1     | 20120101  | 20120112 |   1
  1     | 20120105  | 20120108 |   2

基本上，我希望做的是根据组列产品和价格对数据进行分组更改。

实现这一目标的最简洁方法是什么？

Answer 1

有一种（或多或少）已知的解决此类问题的技术，涉及两个ROW_NUMBER()调用，如下所示：

WITH marked AS (
  SELECT
    *,
    grp = ROW_NUMBER() OVER (PARTITION BY product        ORDER BY date)
        - ROW_NUMBER() OVER (PARTITION BY product, price ORDER BY date)
  FROM #ph
)
SELECT
  product,
  date_from = MIN(date),
  date_to   = MAX(date),
  price
FROM marked
GROUP BY
  product,
  price,
  grp
ORDER BY
  product,
  MIN(date)

输出：

product  date_from   date_to        price 
-------  ----------  -------------  ----- 
1        2012-01-01  2012-01-04     1     
1        2012-01-05  2012-01-08     2     
1        2012-01-09  2012-01-12     1

Answer 2

我是这个论坛的新手，所以希望我的贡献是有帮助的。

如果你真的不想使用CTE（尽管我认为这可能是最好的方法），你可以使用基于集合的代码获得解决方案。您需要测试此代码的性能！。

我添加了一个额外的临时表，以便我可以为每条记录使用唯一的标识符，但我怀疑你的源表中已经有了这个列。所以这是临时表。

    If Exists (SELECT Name FROM tempdb.sys.tables WHERE name LIKE '#phwithId%')
        DROP TABLE #phwithId    

    CREATE TABLE #phwithId
    (
        SaleId INT
        , ProductID INT
        , Price Money
        , SaleDate Date 
    )
    INSERT INTO #phwithId SELECT row_number() over(partition by product order by [date] asc) as SalesId, Product, Price, Date FROM ph

现在是Select语句的主体

    SELECT 
        productId 
        , date_from
        , date_to
        , Price
    FROM
        (   
            SELECT 
                dfr.ProductId
                , ROW_NUMBER() OVER (PARTITION BY ProductId ORDER BY ChangeDate) AS rowno1          
                , ChangeDate AS date_from
                , dfr.Price
            FROM
                (       
                    SELECT
                        sl1.ProductId AS ProductId
                        , sl1.SaleDate AS ChangeDate
                        , sl1.price
                    FROM
                        #phwithId sl1
                    LEFT JOIN
                        #phwithId sl2
                        ON sl1.SaleId = sl2.SaleId + 1
                    WHERE
                        sl1.Price <> sl2.Price OR sl2.Price IS NULL
                ) dfr
        ) da1
    LEFT JOIN
        (   
            SELECT 
                ROW_NUMBER() OVER (PARTITION BY ProductId ORDER BY ChangeDate) AS rowno2
                , ChangeDate AS date_to     
            FROM
                (   
                    SELECT 
                        sl1.ProductId
                        , sl1.SaleDate AS ChangeDate
                    FROM
                        #phwithId sl1
                    LEFT JOIN
                        #phwithId sl3
                        ON sl1.SaleId = sl3.SaleId - 1  
                    WHERE
                        sl1.Price <> sl3.Price OR sl3.Price IS NULL         
                ) dto

        ) da2 
        ON da1.rowno1 = da2.rowno2

通过将数据源偏移量绑定1个记录（+或 - ），我们可以确定价格桶何时发生变化，然后只需将桶的开始和结束日期恢复为单个记录。

一切都有点繁琐，我不确定它会给出更好的表现，但我很享受这个挑战。

Answer 3

WITH marked AS (
  SELECT
    *,
  case
   when (lag(price,1,'') over (partition by product order by date_from)) = price
   then 0 else 1
  end is_price_change
  FROM #ph
),
marked_as_group AS
( SELECT m.*,
       SUM(is_price_change) over (PARTITION BY product order by date_from ROWS 
      BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS price_change_group
  FROM marked m
),
SELECT
  product,
  date_from = MIN(date_from),
  date_to   = MAX(date_to),
  price = MIN(price)
FROM marked_as_group 
GROUP BY
  product,
  price_change_group
ORDER BY
  product,
  date_to

Answer 4

我提出的一个相对“干净”的解决方案是：

;with cte_sort (product, [date], price, [row])
as
    (select product, [date], price, row_number() over(partition by product order by [date] asc) as row
     from #ph)

select a.product, a.[date] as date_from, c.[date] as date_to, a.price 
from cte_sort a
left outer join cte_sort b on a.product = b.product and (a.row+1) = b.row and a.price = b.price
outer apply (select top 1 [date] from cte_sort z where z.product = a.product and z.row > a.row order by z.row) c
where b.row is null
order by a.[date]

我使用了row_number的CTE，因为如果使用dateadd等函数，则无需担心是否缺少任何日期。如果你想拥有date_to列（我这样做），你显然只需要外部申请。

这个解决方案确实解决了我的问题，但是我遇到了一个小问题，让它在我的500万行表上尽可能快地执行。

Answer 5

Create function [dbo].[AF_TableColumns](@table_name nvarchar(55))
returns nvarchar(4000) as
begin
declare @str nvarchar(4000)
    select @str = cast(rtrim(ltrim(column_name)) as nvarchar(500)) + coalesce('         ' + @str , '            ') 
    from information_schema.columns
    where table_name = @table_name
    group by table_name, column_name, ordinal_position 
    order by ordinal_position DESC
return @str
end

--select dbo.AF_TableColumns('YourTable') Select * from YourTable

按顺序更改分组列值对数据进行分组

5 个答案: