SQL 服务器 - SUM 前行，条件链接到原始行

Question

对于以下示例数据集中的每一行，当满足某个条件时，代码会对前 5 行进行求和。

我遇到的问题是条件需要引用原始行评级，例如，只有当评级在当前行的 1 以内时，我才需要对前面的行求和。

示例数据：

DECLARE @tbl TABLE 
             (
                 Team varchar(1),
                 date date, 
                 Rating int, 
                 Score int
             );

INSERT INTO @tbl (Team, Date, Rating, Score)
VALUES
('a', '2020/12/05', '20', '1'),
('a', '2020/12/04', '18', '8'),
('a', '2020/12/03', '21', '3'),
('a', '2020/12/02', '19', '4'),
('a', '2020/12/01', '19', '3');

当前代码：

SELECT
    Rating, 
    SUM(CASE WHEN Rating >= (Rating-1) AND  Rating <= (Rating+1) THEN SCORE END) 
        OVER (partition by Team ORDER BY Date ASC ROWS BETWEEN 5 PRECEDING AND 1 PRECEDING) AS SUM
FROM
    @tbl
ORDER BY 
    Date DESC

Output：

    +------------------+------------+------------+
    |  Rating          | Current    | Required   | 
    +------------------+------------+------------+
    | 20               | 18         |     7      |
    | 18               | 10         |     7      |
    | 21               | 7          |     NULL   |
    | 19               | 3          |     3      |
    | 18               | NULL       |     NULL   |
    +------------------+------------+------------+

问题是代码的以下部分不起作用，因为正在逐行评估评级。

CASE WHEN Rating >= (Rating-1) AND  Rating <= (Rating+1)

我需要它来评估原始行的评级（我已经查看了Top但这不起作用）：

CASE WHEN Rating >= ((SELECT TOP 1 Rating) - 1) AND Rating <= ((SELECT TOP 1 Rating) + 1)

一如既往地感谢任何帮助。

Answer 1

您所描述的听起来像是横向连接：

SELECT t.*, t2.*
FROM @tbl t OUTER APPLY
     (SELECT SUM(t2.score) as score_5
      FROM (SELECT TOP (5) t2.*
            FROM @tbl t2
            WHERE t2.date < t.date
            ORDER BY t2.date DESC
           ) t2
      WHERE t2.rating BETWEEN t.rating - 1 AND t.rating + 1
     ) t2
ORDER BY Date DESC

Answer 2

我不熟悉 sql 服务器语法，但这里是使用 Spark SQL 的方法。 本质上，这个想法是为每对彼此相距 5 行以内的行创建一行，然后sum if 。

select
    Team, date, Rating,
    sum(case when old_score[0] between rating-1 and rating+1 then old_score[1] end) as sum
from (
    select
        *,
        explode_outer(scores) as old_score
    from (
        select
            *,
            collect_list(array(rating, score))
            over (partition by Team order by date rows between 5 preceding and 1 preceding) scores
        from tbl
    )
)
group by Team, date, Rating
order by Team, date, Rating;

这使

a       2020-12-01      19      NULL
a       2020-12-02      19      3
a       2020-12-03      21      NULL
a       2020-12-04      18      7
a       2020-12-05      20      10

并表明您可能在预期的 output 中犯了错误；）