LAG 與 LEAD

Snowflake 的 Window Functions

Jake Roach

Field Data Engineer

LAG

LAG 可讓你把某筆值與「前一筆紀錄」中的值比較

SELECT
    <fields>,


    LAG(<1>, <2>, <3>) OVER(
        PARTITION BY <4>
        ORDER BY <5>
    )


...;

<1>：要取用的前一筆紀錄欄位

<2>：往回「回看」的筆數

<3>：若無該筆紀錄時的預設值，0

<4>：分割（PARTITION）依據的欄位

<5>：決定紀錄排序的欄位

LAG

使用 LAG 產生的範例資料集

LAG

SELECT
    member_id AS m_id, 
    workout_date AS wd,
    calories_burned AS cb,

    -- 取回上一次訓練消耗的卡路里

    LAG(calories_burned, 1) OVER(
        PARTITION BY member_id
        ORDER BY workout_date
    ) AS past_cb,


FROM fitness.workouts;

  m_id  |       wd     |   cb  |  past_cb 
 ------ | ------------ | ----- | ---------
 m_192  |  2024-01-01  |  105  |   null
 m_192  |  2024-01-03  |  156  |   105
 m_192  |  2024-01-04  |   69  |   156
 m_192  |  2024-01-10  |  102  |    69

 m_74   |  2024-02-10  |  374  |   null
 m_74   |  2024-02-13  |  396  |   374
 m_74   |  2024-02-14  |  504  |   396

 m_233  |  2024-03-05  |  51   |   null 
 m_233  |  2024-03-12  |  81   |    51

LAG

SELECT
    ...
    LAG(calories_burned, 1) OVER(
        PARTITION BY member_id
        ORDER BY workout_date
    ) AS past_cb,

    -- 計算本次較上次多（少）消耗的卡路里
``` {sql}
    calories_burned - LAG(calories_burned, 1, calories_burned) OVER(
        PARTITION BY member_id
        ORDER BY workout_date
    ) AS more_cb,


FROM fitness.workouts;

LAG

                  m_id  |       wd     |   cb  |  past_cb  |  more_cb
                ------- | ------------ | ----- | --------- | ---------
                 m_192  |  2024-01-01  |  105  |    null   |      0
                 m_192  |  2024-01-03  |  156  |    105    |     51
                 m_192  |  2024-01-04  |   69  |    156    |    -87
                 m_192  |  2024-01-10  |  102  |     69    |     33

                 m_74   |  2024-02-10  |  374  |    null   |      0
                 m_74   |  2024-02-13  |  396  |    374    |     22
                 m_74   |  2024-02-14  |  504  |    396    |    108

                 m_233  |  2024-03-05  |  51   |    null   |      0
                 m_233  |  2024-03-12  |  81   |     51    |     30

LEAD

LEAD 可讓你把某筆值與「後續」紀錄中的值比較

<1>：要取用的前一筆紀錄欄位

<2>：往前「前看」的筆數

<3>：若無該筆紀錄時的預設值

<4>：分割（PARTITION）依據的欄位

<5>：決定紀錄排序的欄位

SELECT
    <fields>,


    LEAD(<1>, <2>, <3>) OVER(
        PARTITION BY <4>
        ORDER BY <5>
    )


...;

常用於預測型任務

LEAD

SELECT
    member_id, 
    workout_date
    calories_burned,

    -- 在本次訓練之後，找出下一次訓練日期
    LEAD(workout_date, 1) OVER(
        PARTITION BY member_id
        ORDER BY workout_date
    ) AS next_workout_date

FROM fitness.workouts;

LEAD

            m_id  |  workout_date  |  calories_bured  |  next_workout_date
          ------- | -------------- | ---------------- | -------------------
           m_192  |   2024-01-01   |        105       |     2024-01-03
           m_192  |   2024-01-03   |        156       |     2024-01-04
           m_192  |   2024-01-04   |         69       |     2024-01-10
           m_192  |   2024-01-10   |        102       |         null

           m_74   |   2024-02-10   |        374       |     2024-02-13
           m_74   |   2024-02-13   |        396       |     2024-02-14
           m_74   |   2024-02-14   |        504       |         null

           m_233  |   2024-03-05   |         51       |     2024-03-12 
           m_233  |   2024-03-12   |         81       |         null

一起來練習吧！

Snowflake 的 Window Functions