MySQL 根据另一行中最接近的值填充 NULL 行

Question

我正在尝试根据以下逻辑更新scores表中的行：

• 获取没有score的 id 的feat_sum 。
• 对于每个具有feat_sum score的 feat_sum，获取具有最近的feat_sum和score的行，然后将score字段更新为该分数。
• 如果feat_sum差异相同，则选择较小的分数
• id是表的PK

初始表：

scores

| id  | feat_sum | score |
| --- | ---      | ---   |
| 1   | 1.234    | 341   |
| 2   | 5.678    | 758   |
| 3   | 2.234    |  NULL |
| 4   | 8.678    |  NULL |

查询后预计output：

scores

| id  | feat_sum | score |
| --- | ---      | ---   |
| 1   | 1.234    | 341   |
| 2   | 5.678    | 758   |
| 3   | 2.234    | 341   | 
| 4   | 8.678    | 758   |

例如，1.234 比 2.234 更接近 5.678 到 2.234，因此，2.234 的score应该是 341。

我想我在这里得到了基本查询，但我正在努力将最后一点放在一起。

SELECT 
  id,
  feat_sum,
  CASE 
    WHEN score IS NULL
    THEN (SELECT score FROM scores WHERE feat_sum - some_other_feat_sum /* has smallest difference */
  END AS score
FROM scores;

Answer 1

我的思考过程如下，

1. 获取得分为 NULL 的行
2. 计算上面的行与所有其他行的差异，并根据差异、分数对它们进行排名
3. 加入与原始表差异最小的行以获得分数
4. 最后是有条件地显示分数的左连接。

with noscore as 
(
    select * 
    from Table1 
    where score is null
),
alldiff as 
(
    select t1.id, t2.id as diffid, abs(t2.feat_sum-t1.feat_sum) as diff, t2.score 
    from noscore t1 
    inner join Table1 t2 on  
    t1.id != t2.id and t2.score is not null 
    order by diff,score asc
),
diff as 
(
    select *, row_number() over (partition by id order by diff asc) as nr 
    from alldiff
),
mindiff as 
(
    select df.id, t1.feat_sum, t1.score 
    from diff df 
    inner join Table1 t1 on df.diffid = t1.id
    where df.nr = 1
),
updt as 
(
    select t1.id, t1.feat_sum, if(t1.score is null,md.score,t1.score) as score 
    from Table1 t1 
    left join mindiff md on t1.id = md.id
)
update Table1 t1 inner join updt u on t1.id = u.id set t1.score = u.score;

可能有一个 scope 用于查询优化，但我想这暂时有效。

在这里试试

Answer 2

我是 SQL 服务器，我会使用 APPLY 操作，如下所示：

SELECT s.id, s.feat_sum, COALESCE(s.score, alt.score) as score
FROM scores s
CROSS APPLY (
    SELECT TOP 1 score 
    FROM scores s0
    WHERE s0.score IS NOT NULL
    ORDER BY abs(s0.feat_sum - s.feat_sum)
) alt

其他数据库称此为横向连接。 我知道 MySQL 支持这个，但是文档对我来说不是很清楚（它只显示旧的错误的 A,B 语法），所以这可能不太正确：

SELECT s.id, s.feat_sum, COALESCE(s.score, alt.score) as score
FROM scores s
JOIN LATERAL (
    SELECT score 
    FROM scores s0
    WHERE s0.score IS NOT NULL
    ORDER BY abs(s0.feat_sum - s.feat_sum)
    LIMIT 1
) alt

大多数也使用LIMIT 1的横向连接都可以改写为使用窗口 function 来运行得更快。 在这个查询中，我还没有看那么远。

Answer 3

在UPDATE语句中加入一个 CTE 表，该 CTE 通过使用 window function FIRST_VALUE()为每个score为id的最接近score null ：

WITH cte AS (
  SELECT DISTINCT s1.id, 
         FIRST_VALUE(s2.score) OVER (PARTITION BY s1.id ORDER BY ABS(s2.feat_sum - s1.feat_sum), s2.score) AS score
  FROM scores s1 JOIN scores s2
  WHERE s1.score IS NULL AND s2.score IS NOT NULL
)
UPDATE scores s 
INNER JOIN cte c ON c.id = s.id
SET s.score = c.score;

请参阅演示。