PostgreSQL

Question

我有一个 table 这样的：

DATE        ID    ScoreA    ScoreB    ScoreC
20180101    001   91        92        25
20180101    002   81        82        35
20180101    003   71        52        45
20180102    001   82        15        66
20180102    002   69        67        77
...
20180131    003   88        65        73

以一个月的数据为例，我想将它们汇总成一个 MAX 和 MIN 分数的报告，每个分数只有一行 ID。就像：

ID    ScoreA       Date_A              ScoreB        Date_B            ...
001   MAX(ScoreA)  MAX(ScoreA).DATE    MAX(ScoreB)   MAX(ScoreB).DATE  ...
002   MAX(ScoreA)  MAX(ScoreA).DATE    MAX(ScoreB)   MAX(ScoreB).DATE  ...
003   MAX(ScoreA)  MAX(ScoreA).DATE    MAX(ScoreB)   MAX(ScoreB).DATE  ...

其中MAX(ScoreA).DATE表示相应的MAX或MIN分数出现时的DATE（如果MAX分数出现在多个日期，则随机选择一个）

不像常见的合并行的情况，它同时涉及多个列。因为会有 MANY IDs 和 HUNDREDS of Scores （我的意思是有 ScroeA ScroreB ... ScoreZ ... Score1 Score2 ... Score100 ...)，我希望避免使用消耗操作，例如JOINtable。那么有什么好主意吗？

Answer 1

如果您想避免连接，我会提供这样的结构

WITH cte AS (
    SELECT DATE, ID, ScoreA, ScoreB, ScoreC,
        row_number() over (partition by ID order by ScoreA desc) rnA,
        row_number() over (partition by ID order by ScoreB desc) rnB,
        row_number() over (partition by ID order by ScoreC desc) rnC,
    FROM ...
    WHERE DATE BETWEEN ... AND ...
), ids AS (
    SELECT DISTINCT ID FROM cte
)
SELECT ID, 
    (SELECT ScoreA FROM cte t2 WHERE t2.ID = t.ID AND rnA = 1) ScoreA, 
    (SELECT DATE FROM cte t2 WHERE t2.ID = t.ID AND rnA = 1) Date_A,
    (SELECT ScoreB FROM cte t2 WHERE t2.ID = t.ID AND rnB = 1) ScoreB, 
    (SELECT DATE FROM cte t2 WHERE t2.ID = t.ID AND rnB = 1) Date_B,
    (SELECT ScoreC FROM cte t2 WHERE t2.ID = t.ID AND rnC = 1) ScoreC, 
    (SELECT DATE FROM cte t2 WHERE t2.ID = t.ID AND rnC = 1) Date_C
FROM ids t

当您需要 max/min 值的日期或其他属性时，使用行编号而不是聚合函数是合理的：row_number() over (...) as rn 后跟条件 rn = 1

UPD

只要@TaurusDang 想要代码生成，我的解决方案就是让 postgres 完成几乎所有的工作：

WITH cols AS
(
    SELECT column_name
    FROM information_schema.columns
    WHERE table_schema = 'your_schema'
      AND table_name   = 'your_table'
      AND column_name like 'Score%'
)
-- first part: rows for cte subquery
SELECT ',row_number() over (partition by ID order by ' || column_name || ' desc) rn' || column_name
FROM cols
UNION ALL
-- second part: rows for final query
SELECT ',(SELECT ' || column_name || ' FROM cte t2 WHERE t2.ID = t.ID AND rn' || column_name || ' = 1) ' || column_name || ', (SELECT DATE FROM cte t2 WHERE t2.ID = t.ID AND rn' || column_name || ' = 1) Date_' || column_name
FROM cols

只需将生成的行复制到初始查询中：前半部分复制到 cte，下半部分复制到主查询

Answer 2

试试这个

with max_score as
(
    Select    distinct id
        , max(ScoreA) over( partition by id ) as max_ScoreA
        , max(ScoreB) over( partition by id ) as max_ScoreB
        , max(ScoreC) over( partition by id ) as max_Scorec
    from TABLE_NAME
)
Select 
    cte.id
    , max_ScoreA, tbl_a.DATE
    , max_ScoreB, tbl_b.DATE
    , max_ScoreC, tbl_c.DATE
from 
max_score cte
join TABLE_NAME tbl_a
on cte.id = tbl_a.id
and cte.max_ScoreA = tbl_a.ScoreA
join TABLE_NAME tbl_b
on cte.id = tbl_b.id
and cte.max_ScoreB = tbl_b.ScoreB
join TABLE_NAME tbl_c
on cte.id = tbl_c.id
and cte.max_ScoreC = tbl_c.ScoreC
order by 1

Answer 3

这是另一个代码示例，将为您提供所需的所有数据：

select *
from (select
        distinct on (id) id,
        first_value(scorea) over w as a_min,
        last_value(scorea) over w as a_max,
        first_value(date) over w as a_min_d,
        last_value(date) over w as a_max_d
    from the_table
    window w as (partition by id order by scorea)
    order by 1,3 desc) a
join (select
        distinct on (id) id,
        first_value(scoreb) over w as b_min,
        last_value(scoreb) over w as b_max,
        first_value(date) over w as b_min_d,
        last_value(date) over w as b_max_d
    from the_table
    window w as (partition by id order by scoreb)
    order by 1,3 desc) b using(id)
join (select
        distinct on (id) id,
        first_value(scorec) over w as c_min,
        last_value(scorec) over w as c_max,
        first_value(date) over w as c_min_d,
        last_value(date) over w as c_max_d
    from the_table
    window w as (partition by id order by scorec)
    order by 1,3 desc) c using(id)

请注意，有 3 个单独的子查询，每个分数列一个。这里有一些关于 windowing-functions and partitions 的 magic，这对继续阅读很有用。这里有一个棘手的部分是，如果放在同一个查询中，各个分区会相互干扰（至少在我的 pg 9.3.22 上）。

PostgreSQL - 将具有多个属性的多行合并为一行？

PostgreSQL - Combining multiple rows with several attributes into one row?

sql

greatest-n-per-group

greenplum