需要 Gap 和 Island 的最后一个值

Question

我现在有点卡住了，我在这个查询上取得了很好的进展，但我没有得到我期望的结果。我正在使用 BQ 标准 SQL。 Table 看起来像这样。

ID - 帐户的唯一 ID 阶段 - 它在客户生命周期中的位置 CreatedDate - 更新阶段的时间 EndDate - 阶段结束的时间金额 - 该阶段的金额

ID	阶段	创建日期	结束日期	数量
egj3k	发现	2020-12-08	2020-12-08	5000
egj3k	发现	2020-12-08	2020-12-10	6500
egj3k	提案	2020-12-10	2020-12-11	6500
egj3k	提案	2020-12-11	2020-12-15	8000
egj3k	协商	2020-12-15	2020-12-21	7500
egj3k	入职	2020-12-21	2020-12-21	8000
egj3k	入职	2020-12-21	2020-12-22	8000
egj3k	入职	2020-12-21	2020-12-23	10000
egj3k	入职	2020-12-23	2020-12-25	11000
egj3k	收入	2020-12-25	2021-01-31	15000
egj3k	停滞不前	2021-01-31	2021-02-05	7000
egj3k	收入	2021-02-05	2021-03-05	12000

第一个问题是找到每个阶段的最小和最大创建日期，然后显示生命周期，我是这样解决的：

WITH CTE AS
    (
    SELECT  ID, 
            Stage, 
            CreatedDate, 
            EndDate,
            ROW_NUMBER() OVER(PARTITION BY ID ORDER BY CreatedDate)
            - ROW_NUMBER() OVER(PARTITION BY ID, Stage ORDER BY CreatedDate) As WindowId,
            ROW_NUMBER() OVER(PARTITION BY ID ORDER BY CreatedDate DESC)
            - ROW_NUMBER() OVER(PARTITION BY ID, Stage ORDER BY CreatedDate DESC) As ReverseWindowId
    FROM `BQ_TABLE`
    )
    
    SELECT  DISTINCT ID,
            Stage,
            MIN(CreatedDate) OVER(PARTITION BY WindowId, ReverseWindowId) As StartDate,
            NULLIF(MAX(IFNULL(EndDate, '9999-12-31')) OVER(PARTITION BY WindowId, ReverseWindowId), '9999-12-31') As EndDate
    
    FROM CTE 

    where ID = 'egj3k'
    ORDER BY StartDate

产生这个：

ID	阶段	创建日期	结束日期
egj3k	发现	2020-12-08	2020-12-10
egj3k	提案	2020-12-10	2020-12-15
egj3k	协商	2020-12-15	2020-12-21
egj3k	入职	2020-12-21	2020-12-25
egj3k	收入	2020-12-25	2021-01-31
egj3k	停滞不前	2021-01-31	2021-02-05
egj3k	收入	2021-02-05	2021-03-05

完美。但现在我需要找到 window 的最后数量。所以我正在尝试制作这个：

ID	阶段	创建日期	结束日期	数量
egj3k	发现	2020-12-08	2020-12-10	6500
egj3k	提案	2020-12-10	2020-12-15	8000
egj3k	协商	2020-12-15	2020-12-21	7500
egj3k	入职	2020-12-21	2020-12-25	11000
egj3k	收入	2020-12-25	2021-01-31	15000
egj3k	停滞不前	2021-01-31	2021-02-05	7000
egj3k	收入	2021-02-05	2021-03-05	12000

我尝试对 window 和 select 分别进行排名。我也尝试使用 LAST_VALUE 但它没有正确返回。因为您可以看到 Revenue 有两个分类。

rank() OVER (PARTITION BY ID, Stage order by CreatedDate Desc) as Rank_

Where Rank_ = 1

帮帮我？ ;) 有更好的方法吗？

Answer 1

WITH CTE AS
    (
    SELECT  ID, 
            Stage, 
            CreatedDate, 
            EndDate,
            Amount,
            ROW_NUMBER() OVER(PARTITION BY ID ORDER BY CreatedDate)
            - ROW_NUMBER() OVER(PARTITION BY ID, Stage ORDER BY CreatedDate) As WindowId,
            ROW_NUMBER() OVER(PARTITION BY ID ORDER BY CreatedDate DESC)
            - ROW_NUMBER() OVER(PARTITION BY ID, Stage ORDER BY CreatedDate DESC) As ReverseWindowId
    FROM `BQ_Table`
    ),
    CTE_2 as (
    SELECT  ID,
            Stage,
            MIN(CreatedDate) OVER(PARTITION BY WindowId, ReverseWindowId) As StartDate,
            NULLIF(MAX(IFNULL(EndDate, '9999-12-31')) OVER(PARTITION BY WindowId, ReverseWindowId), '9999-12-31') As EndDate,
            Amount,
            row_number() OVER (PARTITION BY  ReverseWindowId, WindowId ORDER BY EndDate desc) as Rev_rank
    
    FROM CTE 


    ORDER BY StartDate)

    SELECT 

    ID,
    Stage,
    StartDate,
    EndDate,
    Amount

    FROM CTE_2 

    Where Rev_rank = 1

它可以工作，但是很丑。仍然有兴趣看看是否有人有更好的主意

Answer 2

您可以使用数组技巧来获取每个岛上的最后一个值。我觉得你的逻辑相当繁琐。我不确定您为什么使用 ROW_NUMBER() 四次，为什么要检查 CreatedDate 的 NULL 值，也不知道为什么使用 SELECT DISTINCT 而不是 GROUP BY.

无论如何，这应该可以满足您的要求：

WITH CTE AS (
      SELECT t.*,
             ROW_NUMBER() OVER (PARTITION BY ID ORDER BY CreatedDate) as seqnum,
             ROW_NUMBER() OVER (PARTITION BY ID, Stage ORDER BY CreatedDate) as seqnum_2
      FROM `BQ_TABLE` t
     )
SELECT ID, Stage, MIN(CreatedDate), MAX(CreatedDate),
       ARRAY_AGG(Amount ORDR BY CreatedDate DESC LIMIT 1)[SAFE_ORDINAL(1)]
FROM cte
GROUP BY ID, Stage, (seqnum - seqnum_2);

Answer 3

考虑以下方法

select id, stage, min(CreatedDate) CreatedDate, max(EndDate) EndDate, 
  array_agg(amount order by EndDate desc limit 1)[offset(0)] amount
from (
  select *, countif(new_stage) over(partition by id order by CreatedDate, EndDate) stage_group
  from (
    select *, 
      ifnull(stage != lag(stage) over(partition by id order by CreatedDate, EndDate), true) new_stage
    from `BQ_Table`    
  )
)
group by id, stage, stage_group

如果应用于您问题中的示例数据 - 输出为

需要 Gap 和 Island 的最后一个值

Need the last value of a Gap and Island

mysql

sql

window-functions

google-bigquery