移动平均与合计

Snowflake 中的 Window Functions

Jake Roach

Field Data Engineer

累计平均与合计

显示"滑动"窗口框架的输出。

"移动"平均与合计

显示移动平均的输出。

"移动"平均与合计

显示移动合计的输出。

移动计算

SELECT
    ...

    AVG(<1>) OVER(
        PARTITION BY ...
        ORDER BY <2>

        -- 窗口覆盖前 X 行
        -- 和后 Y 行记录


        ROWS BETWEEN <X> PRECEDING 
            AND <Y> FOLLOWING

    )
...

指定当前行前后记录数

ROWS BETWEEN X PRECEDING AND Y FOLLOWING

"移动"计算，非累计
仍可用 CURRENT ROW

<X>：向前回看行数

<Y>：向后前瞻行数

创建移动平均

SELECT
    member_id, workout_date, calories_burned,

    -- 使用窗口创建移动平均，包含上一次、当前、下一次训练

    AVG(calories_burned) OVER(
        PARTITION BY member_id
        ORDER BY workout_date

        ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING

    ) AS moving_avg_cb

FROM FITNESS.workouts;

创建移动平均

             member_id  |  workout_date  |   calories_burned  |  moving_avg_cb  
            ----------- | -------------- | ------------------ | --------------- 
               m_192    |   2024-01-01   |         105        |      130.5      
               m_192    |   2024-01-03   |         156        |      110.0      
               m_192    |   2024-01-04   |          69        |      109.0      
               m_192    |   2024-01-10   |         102        |      120.0      
               m_192    |   2024-01-11   |         189        |      127.3     
               m_192    |   2024-01-12   |          91        |      145.0     
               m_192    |   2024-01-16   |         155        |      127.7      
               m_192    |   2024-01-19   |         137        |      133.7      
               m_192    |   2024-01-20   |         109        |      123.0      

               m_74     |   2024-02-10   |         374        |      385.0      
                                        ...

用移动合计通知会员

SELECT
    member_id, workout_date, calories_burned,

    ...

    -- 使用 CURRENT ROW 避免"前瞻"
    SUM(calories_burned) OVER(
        PARTITION BY member_id
        ORDER BY workout_date

        ROWS BETWEEN 2 PRECEDING AND CURRENT ROW

    ) AS last_3_cb

FROM FITNESS.workouts;

用移动合计通知会员

      member_id  |  workout_date  |   calories_burned  |  moving_avg_cb  |  last_3_cb
     ----------- | -------------- | ------------------ | --------------- | -----------
        m_192    |   2024-01-01   |         105        |      130.5      |     105
        m_192    |   2024-01-03   |         156        |      110.0      |     261
        m_192    |   2024-01-04   |          69        |      109.0      |     330
        m_192    |   2024-01-10   |         102        |      120.0      |     327
        m_192    |   2024-01-11   |         189        |      127.3      |     360
        m_192    |   2024-01-12   |          91        |      145.0      |     382
        m_192    |   2024-01-16   |         155        |      127.7      |     435
        m_192    |   2024-01-19   |         137        |      133.7      |     383
        m_192    |   2024-01-20   |         109        |      123.0      |     401

        m_74     |   2024-02-10   |         374        |      385.0      |     374
                                            ...

Passons à la pratique !

Snowflake 中的 Window Functions