PARTITION BY

PostgreSQL 汇总统计与窗口函数

Michel Semaan

Data Scientist

动机

查询

WITH Discus_Gold AS (
  SELECT
    Year, Event, Country AS Champion
  FROM Summer_Medals
  WHERE
    Year IN (2004, 2008, 2012)
    AND Gender = 'Men' AND Medal = 'Gold'
    AND Event IN ('Discus Throw', 'Triple Jump')
    AND Gender = 'Men')

SELECT
  Year, Event, Champion,
  LAG(Champion) OVER
    (ORDER BY Event ASC, Year ASC) AS Last_Champion
FROM Discus_Gold
ORDER BY Event ASC, Year ASC;

结果

| Year | Event        | Champion | Last_Champion |
|------|--------------|----------|---------------|
| 2004 | Discus Throw | LTU      | null          |
| 2008 | Discus Throw | EST      | LTU           |
| 2012 | Discus Throw | GER      | EST           |
| 2004 | Triple Jump  | SWE      | GER           |
| 2008 | Triple Jump  | POR      | SWE           |
| 2012 | Triple Jump  | USA      | POR           |

当 Event 从 Discus Throw 切换到 Triple Jump 时，LAG 拉取的是 Discus Throw 的上一位冠军，而不是 null

引入 PARTITION BY

PARTITION BY 按列的唯一值将表划分为多个分区
- 结果不会汇总到一列
窗口函数对每个分区单独计算
- ROW_NUMBER 会在每个分区重置
- 只有前一行在同一分区时，LAG 才会取到上一值

按单列分区

查询

WITH Discus_Gold AS (...)

SELECT
  Year, Event, Champion,
  LAG(Champion) OVER
    (PARTITION BY Event
     ORDER BY Event ASC, Year ASC) AS Last_Champion
FROM Discus_Gold
ORDER BY Event ASC, Year ASC;

结果

| Year | Event        | Champion | Last_Champion |
|------|--------------|----------|---------------|
| 2004 | Discus Throw | LTU      | null          |
| 2008 | Discus Throw | EST      | LTU           |
| 2012 | Discus Throw | GER      | EST           |
| 2004 | Triple Jump  | SWE      | null          |
| 2008 | Triple Jump  | POR      | SWE           |
| 2012 | Triple Jump  | USA      | POR           |

更复杂的分区

| Year | Country | Event                | Row_N |
|------|---------|----------------------|-------|
| 2008 | CHN     | + 78KG (Heavyweight) | 1     |
| 2008 | CHN     | - 49 KG              | 2     |
| ...  | ...     | ...                  | ...   |
| 2008 | JPN     | 48 - 55KG            | 27    |
| 2008 | JPN     | 48 - 55KG            | 28    |
| ...  | ...     | ...                  | ...   |
| 2012 | CHN     | +75KG                | 32    |
| 2012 | CHN     | - 49 KG              | 33    |
| ...  | ...     | ...                  | ...   |
| 2012 | JPN     | +75KG                | 51    |
| 2012 | JPN     | - 49 KG              | 52    |
| ...  | ...     | ...                  | ...   |

行号应在每个 Year 和 Country 内重置

按多列分区

查询

WITH Country_Gold AS (
  SELECT
    DISTINCT Year, Country, Event
  FROM Summer_Medals
  WHERE
    Year IN (2008, 2012)
    AND Country IN ('CHN', 'JPN')
    AND Gender = 'Women' AND Medal = 'Gold')

SELECT
  Year, Country, Event,
  ROW_NUMBER() OVER (PARTITION BY Year, Country)
FROM Country_Gold;

结果

| Year | Country | Event                | Row_N |
|------|---------|----------------------|-------|
| 2008 | CHN     | + 78KG (Heavyweight) | 1     |
| 2008 | CHN     | - 49 KG              | 2     |
| ...  | ...     | ...                  | ...   |
| 2008 | JPN     | 48 - 55KG            | 1     |
| 2008 | JPN     | 48 - 55KG            | 2     |
| ...  | ...     | ...                  | ...   |
| 2012 | CHN     | +75KG                | 1     |
| 2012 | CHN     | - 49 KG              | 2     |
| ...  | ...     | ...                  | ...   |
| 2012 | JPN     | +75KG                | 1     |
| 2012 | JPN     | - 49 KG              | 2     |
| ...  | ...     | ...                  | ...   |

Passons à la pratique !

PostgreSQL 汇总统计与窗口函数