SQL 服务器 - 按行号分组 - 间隙和孤岛

Question

我有一个 table 有这样的数据

+-------------+--------------+------------+----------------+
| CustomerSID | StartDateSID | EndDateSID | MarketingOptIn |
+-------------+--------------+------------+----------------+
|       12345 |     20101019 |   20131016 | Y              |
|       12345 |     20131017 |   20140413 | Y              |
|       12345 |     20140414 |   20140817 | N              |
|       12345 |     20140818 |   20141228 | N              |
|       12345 |     20141229 |   20150125 | Y              |
|       12345 |     20150126 |          0 | Y              |
+-------------+--------------+------------+----------------+

我需要在此 table 之上创建一个视图，以便将数据格式化为以下格式的标志，基本上是标志为 Y 或 N 的持续时间。（EndDateSID - 0 当前处于活动状态，因此是今天的日期）

+-------------+--------------+------------+----------------+
| CustomerSID | StartDateSID | EndDateSID | MarketingOptIn |
+-------------+--------------+------------+----------------+
|       12345 |     20101019 |   20140413 | Y              |
|       12345 |     20140414 |   20141228 | N              |
|       12345 |     20141229 |   20150825 | Y              |
+-------------+--------------+------------+----------------+

大多数客户的 Flag 仅更改一次，因此以下查询有效：

SELECT 
CH1.CustomerSID
,MIN(CH1.StartDateSID) StartDate
,MAX(ISNULL(NULLIF(CH1.EndDateSID,0),CONVERT(INT, CONVERT(VARCHAR, GETDATE(), 112)))) EndDate
,CH1.MarketingOptIn 
FROM DWH.DimCustomerHistory CH1
GROUP BY CH1.CustomerSID, CH1.MarketingOptIn
ORDER BY CH1.CustomerSID, CH1.MarketingOptIn

我怎样才能为像上面这样的客户实现预期的输出，多次更改标志？

Answer 1

您可以使用以下查询：

SELECT CustomerSID,
       MIN(StartDateSID) AS StartDate,
       MAX(ISNULL(NULLIF(EndDateSID,0),
           CONVERT(INT, CONVERT(VARCHAR, GETDATE(), 112)))) AS EndDate,
       MarketingOptIn
FROM (       
  SELECT CustomerSID, StartDateSID, EndDateSID, MarketingOptIn,
         ROW_NUMBER() OVER (ORDER BY StartDateSID) - 
         ROW_NUMBER() OVER (PARTITION BY CustomerSID, MarketingOptIn 
                            ORDER BY StartDateSID) AS grp     
  FROM DimCustomerHistory ) AS t
GROUP BY CustomerSID, MarketingOptIn, grp
ORDER BY StartDate

计算字段 grp 用于识别具有相同 MarketingOptIn 值的连续条记录。

在外部查询中利用此字段，您可以轻松地 GROUP BY 并以类似于原始查询的方式应用 MIN 和 MAX 聚合函数。

Demo here

Answer 2

这是一个gaps and islands problem。您需要使用 ROW_NUMBER() 来确定您的差距，因此开始阶段将是：

SELECT  CustomerSID,
        StartDateSID,
        EndDateSID,
        MarketingOptIn,
        TotalRowNum = ROW_NUMBER() OVER(PARTITION BY CustomerSID ORDER BY StartDateSID),
        RowNumInGroup = ROW_NUMBER() OVER(PARTITION BY CustomerSID, MarketingOptIn ORDER BY StartDateSID),
        GroupID = ROW_NUMBER() OVER(PARTITION BY CustomerSID ORDER BY StartDateSID) -   
                ROW_NUMBER() OVER(PARTITION BY CustomerSID, MarketingOptIn ORDER BY StartDateSID)
FROM    dbo.YourTable;

输出：

CustomerSID StartDateSID    EndDateSID  MarketingOptIn  TotalRowNum RowNumInGroup   GroupID
---------------------------------------------------------------------------------------------------
12345       20101019        20131016    Y               1           1               0
12345       20131017        20140413    Y               2           2               0
12345       20140414        20140817    N               3           1               2
12345       20140818        20141228    N               4           2               2
12345       20141229        20150125    Y               5           3               2
12345       20150126        0           Y               6           4               2

这里的关键是，通过获取每一行的行号，以及组内每一行的行号，您可以获得一个唯一标识符 (GroupID + MarketingOptIn)，用于标识您的每个岛屿。那么这只是在进行聚合时按此标识符分组的情况：

完整示例

DECLARE @T TABLE 
(   
    CustomerSID INT,
    StartDateSID INT,
    EndDateSID INT,
    MarketingOptIn CHAR(1)
)
INSERT @T 
VALUES
    (12345, 20101019, 20131016, 'Y'),
    (12345, 20131017, 20140413, 'Y'),
    (12345, 20140414, 20140817, 'N'),
    (12345, 20140818, 20141228, 'N'),
    (12345, 20141229, 20150125, 'Y'),
    (12345, 20150126, 0, 'Y');


WITH CTE AS
(
    SELECT  CustomerSID,
            StartDateSID,
            EndDateSID,
            MarketingOptIn,
            GroupID = ROW_NUMBER() OVER(PARTITION BY CustomerSID ORDER BY StartDateSID) -   
                    ROW_NUMBER() OVER(PARTITION BY CustomerSID, MarketingOptIn ORDER BY StartDateSID)
    FROM    @T
)
SELECT  CustomerSID, 
        StartDateSID = MIN(StartDateSID),
        EndDateSID = CASE WHEN MIN(EndDateSID) = 0 THEN CONVERT(INT, CONVERT(VARCHAR(8), GETDATE(), 112)) ELSE MAX(EndDateSID) END,
        MarketingOptIn
FROM    CTE
GROUP BY CustomerSID, MarketingOptIn, GroupID
ORDER BY CustomerSID, StartDateSID;

SQL 服务器 - 按行号分组 - 间隙和孤岛

SQL Server - group by with row number - Gaps and Islands

sql

sql-server

group-by

row-number