如何在 BigQuery 中随机匹配元素

Question

我想通过某种条件随机匹配表的元素（1：1和对称配对，所以如果用户x与用户y匹配，那么用户y与用户x匹配）

WITH tbl AS (
  SELECT 1 AS user, 'A' AS condition
  UNION ALL
  SELECT 2 AS user, 'A' AS condition
  UNION ALL
  SELECT 3 AS user, 'A' AS condition
  UNION ALL
  SELECT 4 AS user, 'B' AS condition
  UNION ALL
  SELECT 5 AS user, 'B' AS condition
  UNION ALL
  SELECT 6 AS user, 'B' AS condition
  UNION ALL
  SELECT 7 AS user, 'B' AS condition
  UNION ALL
  SELECT 8 AS user, 'B' AS condition
  UNION ALL
  SELECT 9 AS user, 'B' AS condition
)
SELECT
  user,
  condition
FROM tbl

如何生成查询以：

1. 随机匹配条件 B 中的所有用户与条件 B 中的另一个用户
2. 对条件 A 执行相同操作，但保留一个条件 A 用户不匹配，因为条件 A 的用户数量为奇数
3. 使用 NULL 匹配标识不匹配的用户（1、2 或 3）

假设结果：

用户	（健康）状况	匹配
1	一个	3
2	一个	NULL
3	一个	1
4	乙	7
5	乙	9
6	乙	8
7	乙	4
8	乙	6
9	乙	5

Answer 1

考虑以下方法

create temp table temp as 
select _0 as user, condition, _1 as match from (
  select user, condition, div(offset, 2) grp, mod(offset, 2) pos
  from (
    select condition, array_agg(user order by rand()) users
    from your_table
    group by condition
  ), unnest(users) user with offset
)
pivot (any_value(user) for pos in (0,1));

select * from (
  select * from temp union all
  select match, condition, user from temp 
)
where not user is null;            

如果应用于您问题中的示例数据 - output 是

如您所见，上面的解决方案需要涉及脚本 - 所以下面是稍微重构的版本，它允许您通过一个“简单”查询获得相同的结果

select * from (
  select if(grp < 0, _0, _1) as user, condition, if(grp < 0, _1, _0) as match from (
    select user, condition, grp, mod(offset, 2) pos
    from (
      select condition, array_agg(user order by rand()) users
      from your_table
      group by condition
    ), unnest(users) user with offset, unnest([div(offset + 2, 2), -1 * div(offset + 2, 2)]) grp
  )
  pivot (any_value(user) for pos in (0,1))
)
where not user is null