从一组重叠的日期范围中查找常见的日期范围

时间：2019-01-17 16:45:35

标签： sql sql-server overlap gaps-and-islands

如何从一组给定的日期范围中查找重叠的（公共）日期范围？

需要考虑特定程序（PID）的所有事件（EID）来找到重叠（通用）日期范围。

示例：程序（PID = 13579）具有两个事件的日期范围（EID = 2）。

以前发布于link

我已经在此处检查（但无用）：Link

Image for Visualizing Overlapped (Common) Date Ranges

示例架构和数据：

CREATE TABLE #EventsTBL
(
    PID INT,
    EID INT,
    StartDate DATETIME,
    EndDate DATETIME
);

INSERT INTO #EventsTBL
VALUES
(13579, '1', '01 Jan 2018', '31 Mar 2019'),
(13579, '2', '01 Feb 2018', '31 May 2018'),
(13579, '2', '01 Jul 2018', '31 Jan 2019'),
(13579, '7', '01 Mar 2018', '31 Mar 2019'),
(13579, '5', '01 Feb 2018', '30 Apr 2018'),
(13579, '5', '01 Oct 2018', '31 Mar 2019'),
(13579, '8', '01 Jan 2018', '30 Apr 2018'),
(13579, '8', '01 Jun 2018', '31 Dec 2018'),
(13579, '13', '01 Jan 2018', '31 Mar 2019'),
(13579, '6', '01 Apr 2018', '31 May 2018'),
(13579, '6', '01 Sep 2018', '30 Nov 2018'),
(13579, '4', '01 Feb 2018', '31 Jan 2019'),
(13579, '19', '01 Mar 2018', '31 Jul 2018'),
(13579, '19', '01 Oct 2018', '28 Feb 2019'),
--
(13570, '16', '01 Feb 2018', '30 Jun 2018'),
(13570, '16', '01 Aug 2018', '31 Aug 2018'),
(13570, '16', '01 Oct 2018', '28 Feb 2019'),
(13570, '23', '01 Mar 2018', '30 Jun 2018'),
(13570, '23', '01 Nov 2018', '31 Jan 2019');

输出应为：

PID     StartDate       EndDate
13579   01-Apr-2018     30-Apr-2018
13579   01-Oct-2018     30-Nov-2018
13570   01-Mar-2018     30-Jun-2018
13570   01-Nov-2018     31-Jan-2019

2 个答案:

答案 0 :(得分：1)

好的，所以我生成了一个CTE，其中包含考虑中的所有日期的日期。

对于每个日期，我都会生成FLAG = 1，我认为我检测到重叠。

然后我将row_number（）用作解决“岛”问题的标准解决方案，并输出flag = 1的“岛”的开始和结束时间

希望对您有所帮助，我会为您提供13570的结果，但是当我理解“重叠”时，整个13579重叠。也许那部分需要进一步的解释，改编。如果您可以根据自己的规则找出如何生成FLAG，则排名部分仍然适用

CREATE TABLE #EventsTBL
(
    PID INT,
    EID INT,
    StartDate DATETIME,
    EndDate DATETIME
);

INSERT INTO #EventsTBL
VALUES
(13579, '1', '01 Jan 2018', '31 Mar 2019'),
(13579, '2', '01 Feb 2018', '31 May 2018'),
(13579, '2', '01 Jul 2018', '31 Jan 2019'),
(13579, '7', '01 Mar 2018', '31 Mar 2019'),
(13579, '5', '01 Feb 2018', '30 Apr 2018'),
(13579, '5', '01 Oct 2018', '31 Mar 2019'),
(13579, '8', '01 Jan 2018', '30 Apr 2018'),
(13579, '8', '01 Jun 2018', '31 Dec 2018'),
(13579, '13', '01 Jan 2018', '31 Mar 2019'),
(13579, '6', '01 Apr 2018', '31 May 2018'),
(13579, '6', '01 Sep 2018', '30 Nov 2018'),
(13579, '4', '01 Feb 2018', '31 Jan 2019'),
(13579, '19', '01 Mar 2018', '31 Jul 2018'),
(13579, '19', '01 Oct 2018', '28 Feb 2019'),
--
(13570, '16', '01 Feb 2018', '30 Jun 2018'),
(13570, '16', '01 Aug 2018', '31 Aug 2018'),
(13570, '16', '01 Oct 2018', '28 Feb 2019'),
(13570, '23', '01 Mar 2018', '30 Jun 2018'),
(13570, '23', '01 Nov 2018', '31 Jan 2019');


SELECT count(enddate) FROM (SELECT CAST('19660423' as date) dt) A LEFT JOIN #EventsTBL B ON A.dt = b.StartDate;

WITH MIN_MAX AS (SELECT MIN(StartDate) S , MAX(EndDate) E FROM #EventsTBL ),
     ALL_DATES AS (SELECT S DT FROM MIN_MAX
                    UNION ALL
                    SELECT DATEADD(day,1,DT) FROM ALL_DATES WHERE DT < (SELECT E FROM MIN_MAX)
                  ),
     BuildFlags AS (SELECT  P.pid,
                            DT,
                            COUNT(e.PID ) CNT, 
                            CASE WHEN COUNT(e.pid) > 1 THEN 1 ELSE 0 END FLAG, 
                            row_number() OVER(partition by p.pid order by DT) RN
                        FROM ALL_DATES A CROSS JOIN (SELECT DISTINCT E2.pid FROM #EventsTBL E2) P
                        LEFT JOIN 
                            #EventsTBL E ON P.PID = E.pid AND
                            A.DT BETWEEN E.StartDate AND E.EndDate GROUP BY P.pid,DT),
    AddRanks AS (SELECT *,rn - row_number()over(partition by pid,flag order by dt) groupRank  FROM BuildFlags)

     select pid,min(dt) as start, max(dt) as ending from AddRanks 
        where flag = 1
        group by pid,grouprank
        order by pid,min(dt)
     option(maxrecursion 0)

编辑-我想我已经明白您的意思了，您想将pid和eid合并为唯一的pid和eid，并附上日期。然后您定义一个重叠，因为所有pid和eid一次都处于活动状态。所以我想出了这个修改

;WITH MIN_MAX AS (SELECT MIN(StartDate) S , MAX(EndDate) E FROM #EventsTBL ),
     ALL_DATES AS (SELECT S DT FROM MIN_MAX
                    UNION ALL
                    SELECT DATEADD(day,1,DT) FROM ALL_DATES WHERE DT < (SELECT E FROM MIN_MAX)
                  ),
     GROUPED AS (SELECT Q.pid,Q.eid,q.dt,case when max(tx.pid) is null then 0 else 1 end YES from (Select * FROM All_Dates cross join (select distinct pid,eid from #EventsTBL) AQ) Q
                                    LEFT JOIN  #EventsTBL TX ON TX.PID = Q.pid and tx.EID = Q.eid and 
                                                Q.DT BETWEEN TX.StartDate AND TX.EndDate GROUP BY q.pid,q.eid,q.dt
                ),                                       
     BuildFlags AS (SELECT g.pid,g.dt, row_number() OVER(partition by g.pid order by g.DT) RN,
          CASE WHEN WQ.tot = (SELECT count(distinct g2.eid)  FROM grouped g2 WHERE g2.PID = G.pid and g2.dt=g.dt and g2.yes=1) then 1 else 0 end FLAG
      FROM GROUPED G cross apply (select count(distinct E9.eid) tot FROM #EventsTBL E9 WHERE E9.PID = G.pid) WQ)
    ,AddRanks AS (SELECT *,rn - row_number()over(partition by pid,flag order by dt) groupRank  FROM BuildFlags)

     select pid,min(dt) as start, max(dt) as ending from AddRanks 
        where flag = 1
        group by pid,grouprank
        order by pid,min(dt)
     option(maxrecursion 0);

答案 1 :(得分：1)

此答案计算重叠间隔的数量。假定具有相同EID的日期范围不重叠。以下是带有解释内联的查询：

DECLARE @EventsTBL TABLE (PID INT, EID INT, StartDate DATETIME, EndDate DATETIME);
INSERT INTO @EventsTBL VALUES
(13579, 1,  '01 Jan 2018', '31 Mar 2019'),
(13579, 2,  '01 Feb 2018', '31 May 2018'),
(13579, 2,  '01 Jul 2018', '31 Jan 2019'),
(13579, 7,  '01 Mar 2018', '31 Mar 2019'),
(13579, 5,  '01 Feb 2018', '30 Apr 2018'),
(13579, 5,  '01 Oct 2018', '31 Mar 2019'),
(13579, 8,  '01 Jan 2018', '30 Apr 2018'),
(13579, 8,  '01 Jun 2018', '31 Dec 2018'),
(13579, 13, '01 Jan 2018', '31 Mar 2019'),
(13579, 6,  '01 Apr 2018', '31 May 2018'),
(13579, 6,  '01 Sep 2018', '30 Nov 2018'),
(13579, 4,  '01 Feb 2018', '31 Jan 2019'),
(13579, 19, '01 Mar 2018', '31 Jul 2018'),
(13579, 19, '01 Oct 2018', '28 Feb 2019'),
(13570, 16, '01 Feb 2018', '30 Jun 2018'),
(13570, 16, '01 Aug 2018', '31 Aug 2018'),
(13570, 16, '01 Oct 2018', '28 Feb 2019'),
(13570, 23, '01 Mar 2018', '30 Jun 2018'),
(13570, 23, '01 Nov 2018', '31 Jan 2019');

WITH cte1 AS (
    /*
     * augment the data with the number of distinct EID per PID
     * we will need this later
     */
    SELECT e.PID, a.EIDCount, StartDate, EndDate
    FROM @EventsTBL AS e
    JOIN (
        SELECT PID, COUNT(DISTINCT EID) AS EIDCount
        FROM @EventsTBL
        GROUP BY PID
    ) AS a ON e.PID = a.PID
), cte2 AS (
    /*
     * build a list of "points in time" at which an event started or ended
     * and the number concurrent events changed
     * the zero value rows are required!
     */
    SELECT PID, EIDCount, StartDate AS pdate, 1 AS pval
    FROM cte1
    UNION ALL
    SELECT PID, EIDCount, EndDate, 0
    FROM cte1
    UNION ALL
    SELECT PID, EIDCount , DATEADD(DAY, 1, EndDate), -1
    FROM cte1
), cte3 AS (
    /*
     * calculate running sum of pval over dates; minus ones first
     */
    SELECT PID, EIDCount, pdate, SUM(pval) OVER (PARTITION BY PID ORDER BY pdate, pval) AS evtcount
    FROM cte2
), cte4 AS (
    /*
     * consolidate data for same dates and we are done with the main part
     */
    SELECT PID, EIDCount, pdate, MAX(evtcount) AS evtcount
    FROM cte3
    GROUP BY PID, EIDCount, pdate
), cte5 AS (
    /*
     * assign "change flag" to rows where number of concurrent events
     * enters or exits the required count w.r.t. previous row
     */
    SELECT PID, EIDCount, pdate, evtcount, CASE
        WHEN evtcount < EIDCount AND LAG(evtcount) OVER (PARTITION BY PID ORDER BY pdate) < EIDCount THEN 0
        WHEN evtcount = EIDCount AND LAG(evtcount) OVER (PARTITION BY PID ORDER BY pdate) = EIDCount THEN 0
        ELSE 1
    END AS chg
    FROM cte4
), cte6 AS (
    /*
     * convert "change flag" to "group numbers" over consecutive rows using running sum
     */
    SELECT PID, EIDCount, pdate, evtcount, SUM(chg) OVER (PARTITION BY PID ORDER BY pdate) AS grp
    FROM cte5
)
/*
 * group rows by pid and group numbers
 */
SELECT PID, MIN(pdate) AS StartDate, MAX(pdate) AS EndDate
FROM cte6
WHERE evtcount = EIDCount
GROUP BY PID, grp
ORDER BY PID, StartDate

Demo on db<>fiddle

相关问题
最新问题