将 n 列一起发生的事件转换为两列成对的事件

Posted 2023-03-24

【中文标题】将 n 列一起发生的事件转换为两列成对的事件

【英文标题】：Transform n columns of events occurring together into two columns of pairs of occurrences

【发布时间】：2021-08-06 09:14:35

【问题描述】：

我目前有一张这样的桌子：

A	B	C	D	n
1	1	0	0	50
0	0	1	0	100
0	1	1	1	200

其中第一行表示事件 A 和 B 一起发生了 50 次，第二行表示事件 C 本身发生了 100 次，依此类推。

实际上，我有大约 10 个事件和 10 个事件允许的尽可能多的组合，我想将表格转换为 3 列，其中上表的相关行如下所示

Event1	Event2	n
A	B	50
C	C	100
B	C	200
B	D	200
C	D	200

我不知道从哪里解决这个问题。我想查看 SQL 分解函数，或旋转数据，但不知道如何处理这些数据。

【问题讨论】：

表中是否有主键，例如 id？

@forpas 该表派生自具有主键的表 - 列标题（事件）都是主键。此表中没有主键，仅用于计数/分析目的

为什么省略了 A->A、B->B 和 D->D？

@berihulel 正如我在帖子中提到的，我只在顶部包含了与表格相关的行。 A、B 或 D 都不是自己发生的。

@15150776 。 . .为什么在第二个示例中需要 C/C 而在第一个示例中不需要 A/A？

【参考方案1】：

有很多方法可以做到这一点。只需将UNION 术语添加到data2 CTE 术语，就可以扩展以处理更多列。我敢肯定有办法缩短它，但这是比较清楚的。

我在这个测试用例中使用了 mysql 8.0.26。

注意：MariaDB 10.5.0 在这方面存在一些错误。如果您碰巧尝试了 MariaDB，可能需要进行一些编辑来解决这些错误。没什么可怕的。只是烦人。

data0 提供您的初始表格。 data1 只是强制对该数据进行排序，然后为每行分配一个唯一的 ID data2 通过UNION 部分规范化数据。添加术语以处理更多列。 data 在每个 id 中添加一个 COUNT (cnt) 以标识没有配对的事件。 tcross 生成对的部分笛卡尔积，防止反射等。

WITH data0 (a, b, c, d, n) AS (
         SELECT 1, 1, 0, 0,  50 UNION
         SELECT 0, 0, 1, 0, 100 UNION
         SELECT 0, 1, 1, 1, 200
     )
   , data1 (a, b, c, d, n, id) AS (
         SELECT t.*, ROW_NUMBER() OVER (ORDER BY n) FROM data0 AS t
     )
   , data2 (event, n, id) AS (
         SELECT 'A', n, id FROM data1 WHERE a = 1 UNION
         SELECT 'B', n, id FROM data1 WHERE b = 1 UNION
         SELECT 'C', n, id FROM data1 WHERE c = 1 UNION
         SELECT 'D', n, id FROM data1 WHERE d = 1
     )
   , data (event, n, id, cnt) AS (
         SELECT t.*, COUNT(*) OVER (PARTITION BY id) FROM data2 AS t
     )
   , tcross (event1, event2, n, id) AS (
         SELECT t1.event, COALESCE(t2.event, t1.event), t1.n, t1.id
           FROM      data AS t1
           LEFT JOIN data AS t2
             ON t1.id = t2.id
            AND t1.event < t2.event
          WHERE t2.event IS NOT NULL OR t1.cnt = 1
     )
SELECT event1, event2, n
  FROM tcross
 ORDER BY id, event1, event2
;

结果：

+--------+--------+-----+
| event1 | event2 | n   |
+--------+--------+-----+
| A      | B      |  50 |
| C      | C      | 100 |
| B      | C      | 200 |
| B      | D      | 200 |
| C      | D      | 200 |
+--------+--------+-----+

Fully working test case

【参考方案2】：

您可以像这样对union all 使用蛮力方法：

select 'A' as event1, 'B' as event2, n
from t
where a = 1 and b = 1
union all
select 'A' as event1, 'C' as event2, n
from t
where a = 1 and c = 1
union all
select 'A' as event1, 'D' as event2, n
from t
where a = 1 and d = 1
union all
select 'B' as event1, 'C' as event2, n
from t
where b = 1 and c = 1
union all
select 'B' as event1, 'D' as event2, n
from t
where b = 1 and d = 1
union all
select 'C' as event1, 'D' as event2, n
from t
where c = 1 and d = 1;

您也可以使用join 方法：

select x.event1, x.event2, t.n
from t join
     (select 1 as a, 1 as b, 0 as c, 0 as d, 'A' as event1, 'B' as event2 union all
      select 1 as a, 0 as b, 1 as c, 0 as d, 'A' as event1, 'C' as event2 union all
      select 1 as a, 0 as b, 0 as c, 1 as d, 'A' as event1, 'D' as event2 union all
      select 0 as a, 1 as b, 1 as c, 0 as d, 'B' as event1, 'C' as event2 union all
      select 0 as a, 1 as b, 0 as c, 1 as d, 'B' as event1, 'D' as event2 union all
      select 0 as a, 0 as b, 1 as c, 1 as d, 'C' as event1, 'D' as event2
     ) x
     on (x.a = t.a or x.a = 0) and
        (x.b = t.b or x.b = 0) and
        (x.c = t.c or x.c = 0) and
        (x.d = t.d or x.d = 0);

  

【讨论】：

这不会返回预期结果的第二行。