如何识别重叠行答案

【问题标题】：How to identify overlapping rows如何识别重叠行
【发布时间】：2020-12-27 05:06:21
【问题描述】：

我需要识别重叠的行。下面是表格，我需要列重叠：

| identifier    | status    | startDate     | endDate       | pID   | OVERLAPPING   |
|------------   |---------- |------------   |------------   |-----  |-------------  |
| A             | Approved  | 2020-10-01    | 2020-10-07    | x1    | Yes           |
| A             | Approved  | 2020-10-01    | 2020-10-07    | x2    | No            |
| A             | Approved  | 2020-10-01    | 2020-10-07    | x3    | Yes           |
| A             | Approved  | 2020-10-01    | 2020-10-07    | x4    | No            |
| B             | Approved  | 2020-10-10    | 2020-10-12    | x2    | No            |
| B             | Approved  | 2020-10-10    | 2020-10-12    | x5    | No            |
| C             | Rejected  | 2020-10-05    | 2020-10-06    | x3    | No            |
| C             | Rejected  | 2020-10-05    | 2020-10-06    | x7    | No            |
| C             | Rejected  | 2020-10-05    | 2020-10-06    | x8    | No            |
| C             | Rejected  | 2020-10-05    | 2020-10-06    | x9    | No            |
| D             | Approved  | 2020-10-01    | 2020-10-07    | x5    | No            |
| D             | Approved  | 2020-10-01    | 2020-10-07    | x1    | Yes           |
| E             | Approved  | 2020-10-03    | 2020-10-04    | x3    | Yes           |

由于 C 处于拒绝状态，因此任何具有 C 标识符的内容都不会被计入查找重叠行。因此即使 C 中的坚韧 x3 与 A 中的 x3 重叠，也不会被视为重叠。
x1 在 A 和 D 中重叠。因此这两行都将被计为重叠。
A 中的 x3 与 E 中的 x3 重叠，因为 E 的开始日期和结束日期在 A 的周期内。
B 和 D 中的 x5 不重叠，因为 B 和 D 的日期不重叠。

如果开始日期和结束日期相同，我可以通过创建一个由开始日期、结束日期和 pID 组合而成的列，然后计算所有行的该字段来获得此信息。如果它大于 1，那么我标记重叠。但这不包括 x3 的场景，其中开始日期和日期不相同但在一段时间内仍然重叠。

【问题讨论】：

Redshift 还是 Postgres？虽然它们有一些古老的根源，但它们是截然不同的产品

标签： sql date amazon-redshift

【解决方案1】：

这样的？

（如果一行与其他行重叠，我使用相关子查询来避免连接返回多行。）

SELECT
  *,
  (
    SELECT 1
      FROM yourTable AS lookup
     WHERE lookup.identifier <> yourTable.identifier  -- Don't check overlaps with itself
       AND lookup.pID         = yourTable.pID
       AND lookup.startDate  <= yourTable.endDate
       AND lookup.endDate    >= yourTable.startDate
     LIMIT 1
  )
FROM
  yourTable

注意>= 和<=，这取决于您使用的是包容性（我希望不是）还是排他性（我希望如此）endDate。

【讨论】：

使用CASE WHEN EXISTS(...) THEN 1 ELSE 0 END 可能比SELECT 1 ... LIMIT 1 更有效。
@GMB - 我认为 RedShift 只允许 EXISTS() 并返回布尔类型。
是的，打字更短:)

【解决方案2】：

根据你的叙述，这似乎符合你的逻辑：

select *,
   case 
     when status = 'Rejected' then 'No'
       -- previous row overlaps
     when startDate < -- maybe <=
          max(case when status <> 'Rejected' then endDate end)
          over (partition by pID
                order by startDate, endDate desc
                rows between unbounded preceding and 1 preceding)
       -- next row overlaps
       or endDate > -- maybe >=
          min(case when status <> 'Rejected' then startDate end)
          over (partition by pID
                order by startDate, endDate desc
                rows between 1 following and unbounded following)
     then 'Yes'
     else 'No'
   end 
from tab

【讨论】：

【解决方案3】：

试试这个（虽然我使用了一个布尔值与真假...

WITH
input(identifier,status,startDate,endDate,pID,OVERLAPPING) AS (
          SELECT 'A','Approved',DATE '2020-10-01',DATE '2020-10-07','x1','Yes'
UNION ALL SELECT 'A','Approved',DATE '2020-10-01',DATE '2020-10-07','x2','No'
UNION ALL SELECT 'A','Approved',DATE '2020-10-01',DATE '2020-10-07','x3','Yes'
UNION ALL SELECT 'A','Approved',DATE '2020-10-01',DATE '2020-10-07','x4','No'
UNION ALL SELECT 'B','Approved',DATE '2020-10-10',DATE '2020-10-12','x2','No'
UNION ALL SELECT 'B','Approved',DATE '2020-10-10',DATE '2020-10-12','x5','No'
UNION ALL SELECT 'C','Rejected',DATE '2020-10-05',DATE '2020-10-06','x3','No'
UNION ALL SELECT 'C','Rejected',DATE '2020-10-05',DATE '2020-10-06','x7','No'
UNION ALL SELECT 'C','Rejected',DATE '2020-10-05',DATE '2020-10-06','x8','No'
UNION ALL SELECT 'C','Rejected',DATE '2020-10-05',DATE '2020-10-06','x9','No'
UNION ALL SELECT 'D','Approved',DATE '2020-10-01',DATE '2020-10-07','x5','No'
UNION ALL SELECT 'D','Approved',DATE '2020-10-01',DATE '2020-10-07','x1','Yes'
UNION ALL SELECT 'E','Approved',DATE '2020-10-03',DATE '2020-10-04','x3','Yes'
)
SELECT 
  *
, status = 'Approved'
  AND (
     COALESCE(LAG(enddate)    OVER(w) ,'0001-01-01')> startdate
  OR COALESCE(LEAD(startdate) OVER(w) ,'9999-12-31')< enddate
  ) AS overlap
FROM input
WINDOW w AS (PARTITION BY pid ORDER BY startdate)
ORDER BY
  identifier
, startdate
;
-- out identifier|status  |startDate |endDate   |pID|OVERLAPPING|overlap
-- out A         |Approved|2020-10-01|2020-10-07|x3 |Yes        |true
-- out A         |Approved|2020-10-01|2020-10-07|x2 |No         |false
-- out A         |Approved|2020-10-01|2020-10-07|x4 |No         |false
-- out A         |Approved|2020-10-01|2020-10-07|x1 |Yes        |true
-- out B         |Approved|2020-10-10|2020-10-12|x5 |No         |false
-- out B         |Approved|2020-10-10|2020-10-12|x2 |No         |false
-- out C         |Rejected|2020-10-05|2020-10-06|x3 |No         |false
-- out C         |Rejected|2020-10-05|2020-10-06|x8 |No         |false
-- out C         |Rejected|2020-10-05|2020-10-06|x7 |No         |false
-- out C         |Rejected|2020-10-05|2020-10-06|x9 |No         |false
-- out D         |Approved|2020-10-01|2020-10-07|x1 |Yes        |true
-- out D         |Approved|2020-10-01|2020-10-07|x5 |No         |false
-- out E         |Approved|2020-10-03|2020-10-04|x3 |Yes        |true

【讨论】：