添加具有滚动计算分组依据的列

Question

我有一个 table 这样的：

current_date	user_id	mode_name	mode_time
2021-10-01	1	game	10
2021-10-02	1	game	10
2021-10-02	1	tv	30
2021-10-09	1	music	10
2021-10-15	1	music	40
2021-10-01	2	music	10
2021-10-01	2	game	10
2021-10-04	2	game	10
2021-10-04	2	music	20
2021-10-05	2	tv	40
2021-10-11	2	tv	40
2021-10-12	2	game	20

我想添加两列：

1. 最喜欢的列 mode_name，根据每个 user_id
的 mode_time 列的累计总和
2. 每个 user_id
收藏夹 mode_name 中 mode_time 列的累计总和的列

所需的 table 应如下所示：

current_date	user_id	mode_name	mode_time	favourite_mode	favourite_mode_time
2021-10-01	1	game	10	game	10
2021-10-02	1	game	10	tv	30
2021-10-02	1	tv	30	tv	30
2021-10-09	1	music	10	tv	30
2021-10-15	1	music	40	music	50
2021-10-01	2	music	10	game	10
2021-10-01	2	game	10	game	10
2021-10-04	2	game	10	music	30
2021-10-04	2	music	20	music	30
2021-10-05	2	tv	40	tv	40
2021-10-11	2	tv	40	tv	80
2021-10-12	2	game	20	tv	80

Table 可以在这里找到 https://dbfiddle.uk/?rdbms=sqlserver_2019&fiddle=e05302a2cfd81a2a55de811e294f513e

Answer 1

您可以使用 max 与用户和模式分区来计算模式的滚动总和，然后使用 max 和 max_by 在外部 select 中获取相应的值:

-- sample data
WITH dataset (date, user_id, mode_name, mode_time) AS (
    values ('2021-10-01', 1, 'game', 10),
        ('2021-10-02', 1, 'game', 10),
        ('2021-10-02', 1, 'tv', 30),
        ('2021-10-09', 1, 'music', 10),
        ('2021-10-15', 1, 'music', 40),
        ('2021-10-01', 2, 'game', 10),
        ('2021-10-01', 2, 'music', 10),
        ('2021-10-04', 2, 'game', 10),
        ('2021-10-04', 2, 'music', 20),
        ('2021-10-05', 2, 'tv', 40),
        ('2021-10-11', 2, 'tv', 40),
        ('2021-10-12', 2, 'game', 20)
) 

--query
SELECT date, user_id, mode_name, mode_time,
    max_by(mode_name, mode_time_rolling_time) OVER (
        PARTITION BY user_id
        ORDER BY date
    ) AS favourite_mode,
    max(mode_time_rolling_time) OVER (
        PARTITION BY user_id
        ORDER BY date
    ) AS favourite_mode_time
FROM(
        SELECT *,
            sum(mode_time) OVER (
                PARTITION BY user_id,
                mode_name
                ORDER BY date
            ) AS mode_time_rolling_time
        FROM dataset
    )
ORDER BY user_id, date

输出：

date	user_id	mode_name	mode_time	favourite_mode	favourite_mode_time
2021-10-01	1	game	10	game	10
2021-10-02	1	game	10	tv	30
2021-10-02	1	tv	30	tv	30
2021-10-09	1	music	10	tv	30
2021-10-15	1	music	40	music	50
2021-10-01	2	game	10	game	10
2021-10-01	2	music	10	game	10
2021-10-04	2	music	20	music	30
2021-10-04	2	game	10	music	30
2021-10-05	2	tv	40	tv	40
2021-10-11	2	tv	40	tv	80
2021-10-12	2	game	20	tv	80