Question

基本上，我想逐个对我的表进行分区，并在该窗口中最近一个日期的n个月窗口内为每条记录添加一个行号（EVENT）。对于下面的例子，n = 3。

示例数据：

PERSON  DATE (yyyy-mm-dd)
A       2014-05-02  
A       2014-01-09
A       2014-01-08
A       2014-01-07
A       2014-01-02
B       2014-07-11
B       2014-06-12
B       2014-01-10
C       2014-11-11

结果：

PERSON  DATE (yyyy-mm-dd)   EVENT   
A       2014-05-02          1       
A       2014-01-09          2       
A       2014-01-08          2       
A       2014-01-07          2       
A       2014-01-02          2       
B       2014-07-11          1       
B       2014-06-12          1       
B       2014-01-10          2       
C       2014-11-11          1

我如何获得这些结果？

我一直试图通过递归CTE解决这个问题，但递归步骤让我失望：

WITH testCTE (PERSON, DATE, EVENT)
AS
(
    SELECT A.PERSON, A.DATE, 1 AS EVENT
    FROM [dbo].[Records] A JOIN (SELECT MAX(PERSON) AS PERSON, MAX(DATE) AS DATE FROM [dbo].[Records] GROUP BY PERSON) B
        ON A.PERSON = B.PERSON AND A.DATE >= DATEADD(MONTH, -3, B.DATE)

    UNION ALL

    -- Not sure what to put here. This gives an error:
    -- Recursive references are not allowed on the right hand side of an EXCEPT operator in the recursive part of recursive CTEs.
    (
        SELECT PERSON, DATE, EVENT+1 AS EVENT
        FROM [dbo].[Records]

        EXCEPT

        SELECT A.PERSON, A.DATE, EVENT
        FROM [dbo].[Records] A JOIN testCTE B
            ON A.PERSON = B.PERSON AND A.DATE = B.DATE
     )
)
    SELECT * FROM testCTE

我目前正在使用SQL Server 2008，但最终将在Oracle 10g中实现。

Answer 1

这可能对您有用 - Oracle解决方案非常相似：

WITH x1 AS (
  SELECT person, MAX(dt) AS max_dt
    FROM person_event
   GROUP BY person
)
SELECT p1.person, p1.dt, FLOOR(DATEDIFF(month, p1.dt, x1.max_dt)/3) + 1
  FROM person_event p1 INNER JOIN x1
    ON p1.person = x1.person

Please see SQL Fiddle Demo here

在Oracle中，您可以执行以下操作：

WITH x1 AS (
  SELECT person, MAX(dt) AS max_dt
    FROM person_event
   GROUP BY person
)
SELECT p1.person, p1.dt, TRUNC(MONTHS_BETWEEN(p1.dt, x1.max_dt)/3) + 1
  FROM person_event p1 INNER JOIN x1
    ON p1.person = x1.person

Answer 2

ROW_NUMBER（）的另一个例子：

with a AS(SELECT *, ROW_NUMBER() OVER (PARTITION BY PERSON ORDER BY DT DESC) AS rn FROM #Records),
c AS(
  SELECT PERSON AS PERSON, DT, DT AS eventStart, 1 AS EVENT, rn 
  FROM a WHERE rn=1
  UNION ALL
  SELECT c.PERSON, r.DT, 
    CASE WHEN r.DT < DATEADD(MONTH, -3, c.eventStart) THEN r.DT ELSE c.eventStart END,
    CASE WHEN r.DT < DATEADD(MONTH, -3, c.eventStart) THEN c.EVENT + 1 ELSE c.EVENT END,
    r.rn
  FROM c INNER JOIN a r ON c.PERSON = r.PERSON and c.rn=r.rn-1
  )
SELECT PERSON, DT, EVENT FROM c order by 1, 2 desc;

Answer 3

寻找有3个月差距的地方。这些开始时段，然后使用此信息分配组。

with r as (
      select r.*,
             (case when r.date > dateadd(month, 3, rprev.date) then 1 else 0 end) as IsSeqStart
      from records r outer apply
           (select top 1 r2.*
            from records r2
            where r2.person = r.person and
                  r2.date < r.date
            order by r2.date desc
           ) prevr
     )
select r.*, r2.event
from r outer apply
     (select count(*) as event
      from records r2
      where r2.person = r.person and
            r2.date < r.date
     ) r2;

使用lag()和累积和更容易表达逻辑，但这些不可用。

分区和滑动窗口上的行号

3 个答案: