sql - 按分组列值的变化顺序分组数据

Question

有以下数据

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（请参阅日期重叠）。

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

所以基本上我想要做的是根据组列产品和价格对数据的步进变化进行分组。

实现这一目标的最干净的方法是什么？

score 32 · Accepted Answer

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

score 2 · Accepted Answer

我是这个论坛的新手，所以希望我的贡献是有帮助的。

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

我已经添加了一个额外的临时表，以便我可以为每条记录使用一个唯一标识符，但我怀疑你的源表中已经有了这个列。所以这里是临时表。

    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 条记录（+或-），我们可以识别价格桶何时发生变化，然后只需将桶的开始和结束日期返回到单个记录中即可。

有点繁琐，我不确定它是否会提供更好的性能，但我喜欢挑战。

score 2 · Accepted Answer

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

score 0 · Accepted Answer

我提出的一种相对“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 CTE，row_number因为如果您使用诸如dateadd. 如果您想拥有 date_to 列（我这样做），您显然只需要外部应用。

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。

score -1 · Accepted Answer

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

sql - 按分组列值的变化顺序分组数据

5 回答 5

Related

Reference