SQL Server 滞后/跨组领先/密集排名

Question

我希望找到有关如何填写超前值和滞后值的答案。

输入：

BRAND   Promo_Start VALUE
TESLA   2016-06-05  NULL
TESLA   2016-06-12  NULL
TESLA   2016-06-19  40000
TESLA   2016-06-26  75000
TESLA   2016-07-03  75000
TESLA   2016-07-10  NULL

我想要的输出是：

BRAND   Promo_Start VALUE
TESLA   2016-06-05  40000
TESLA   2016-06-12  40000
TESLA   2016-06-19  40000
TESLA   2016-06-26  75000
TESLA   2016-07-03  75000
TESLA   2016-07-10  75000

我已经能够填写最后一个值，但是，我没有成功填写前两个值。

BRAND   Promo_Start VALUE   FILLED_VALUE
TESLA   2016-06-05  NULL    NULL
TESLA   2016-06-12  NULL    NULL
TESLA   2016-06-19  40000   40000
TESLA   2016-06-26  75000   75000
TESLA   2016-07-03  75000   75000
TESLA   2016-07-10  NULL    75000

使用这个查询：

WITH help1 as (
SELECT *,
CASE WHEN [VALUE] IS NULL THEN 0 ELSE 1 END CHANGEINDICATOR
FROM #SOExample
)
, help2 as (
SELECT *, SUM(CHANGEINDICATOR) OVER (ORDER BY [Promo_Start]) RowGroup from help1
)
SELECT [BRAND],[Promo_Start],[VALUE],
CASE WHEN [VALUE] IS NOT NULL THEN [VALUE]
ELSE FIRST_VALUE([VALUE]) OVER (PARTITION BY RowGroup ORDER BY [Promo_Start])
END [FILLED_VALUE]
FROM help2
GO

表是使用以下方法创建的：

CREATE TABLE #SOExample
    ([BRAND] varchar(10),[Promo_Start] varchar(10), [VALUE] varchar(15))
;
GO

INSERT INTO #SOExample
    ([BRAND],[Promo_Start], [VALUE])
VALUES
    ('TESLA', '2016-06-05',NULL),
    ('TESLA', '2016-06-12',NULL),
    ('TESLA', '2016-06-19','40000'),
    ('TESLA', '2016-06-26','75000'),
    ('TESLA', '2016-07-03','75000'),
    ('TESLA', '2016-07-10',NULL)
;
GO

我认为这个问题可能类似于：LAG() / LEAD() of the next rank (Postgresql) 我查看了 this 创建标志和 this 因为这似乎是一个类似的问题。

我还研究了 dense_rank 并使用了更改指示器（从 NULL 更改为下一行中的值，并将值更改为下一行中的 NULL）。

Answer 1

你只需要处理RowGroup = 0时的情况

WITH help1 as (
SELECT *,
CASE WHEN [VALUE] IS NULL THEN 0 ELSE 1 END CHANGEINDICATOR
FROM #SOExample
)
, help2 as (
SELECT *, SUM(CHANGEINDICATOR) OVER (ORDER BY [Promo_Start]) RowGroup from help1
)
SELECT [BRAND],[Promo_Start],[VALUE],
CASE WHEN RowGroup = 0 THEN (Select [value] from help2 where RowGroup = 1)
ELSE FIRST_VALUE([VALUE]) OVER (PARTITION BY RowGroup ORDER BY [Promo_Start])
END [FILLED_VALUE]
FROM help2

Answer 2

有2 OUTER APPLY，第一个获取下一个非空值，第二个获取上一个非空值。

SELECT  *, NEW_VALUE = COALESCE ( so.VALUE, n.FILLED_VALUE, p.FILLED_VALUE)
FROM    #SOExample so
        OUTER APPLY
        (
            SELECT  TOP 1 FILLED_VALUE = x.VALUE
            FROM    #SOExample x
            WHERE   x.BRAND     = so.BRAND
            AND     x.Promo_Start   >= so.Promo_Start
            AND     x.VALUE     IS NOT NULL
            ORDER BY x.Promo_Start
        ) n
        OUTER APPLY
        (
            SELECT  TOP 1 FILLED_VALUE = x.VALUE
            FROM    #SOExample x
            WHERE   x.BRAND         = so.BRAND
            AND     x.Promo_Start   <= so.Promo_Start
            AND     x.VALUE     IS NOT NULL
            ORDER BY x.Promo_Start DESC
        ) p