sql - 如何使用 SQL 根据事件类型对事件进行顺序编号？

Question

我试图弄清楚每种“事件类型”发生的天数。例如，我有几个用户、几个事件类型和日期。我想为“自上次事件以来的天数”添加一列（见图）。这个（特别是雪花）的 SQL 语法是什么？该图像准确地显示了我正在尝试做的事情，但我在 Excel 中创建了该示例。

输出目标

Answer 1

在评论“如果事件类型没有改变，取计数+1，如果事件类型改变，从1开始计数”，这是重要的陈述，可以翻译为事件类型的窗口函数，然后使用row_number功能。

有一个有趣的窗口函数，称为conditional_change_event https://docs.snowflake.com/en/sql-reference/functions/conditional_change_event.html ，它有助于当前行与前一行的变化。

请注意，有一个order by子句conditional_change_event让我们引入一个autoincrementid 列。也就是说，让我们开始使用一些 sql 查询

使用示例值创建表：

create or replace temporary table _temp (
 id int autoincrement,
 user int,
 _type varchar,
 dates date
);

insert into _temp(user, _type, dates)
values (12345,'active',to_date('1/15/21', 'MM/DD/YY')),
(12345,'active',to_date('1/16/21', 'MM/DD/YY')),
(12345,'active',to_date('1/17/21', 'MM/DD/YY')),
(12345,'dormant',to_date('1/18/21', 'MM/DD/YY')),
(12345,'dormant',to_date('1/19/21', 'MM/DD/YY')),
(12345,'churned',to_date('1/20/21', 'MM/DD/YY')),
(12345,'churned',to_date('1/21/21', 'MM/DD/YY')),
(12345,'churned',to_date('1/22/21', 'MM/DD/YY')),
(39498,'active',to_date('1/15/21', 'MM/DD/YY')),
(39498,'active',to_date('1/16/21', 'MM/DD/YY')),
(39498,'dormant',to_date('1/17/21', 'MM/DD/YY')),
(39498,'churned',to_date('1/18/21', 'MM/DD/YY'));

目标

with event_changes as (
  select *,
  conditional_change_event(_type) over (order by id) as type_changes
  from _temp
)
select *, 
row_number() over(partition by _type, type_changes order by id) as days_since_last_event
from event_changes
order by id;

输出

  ID    USER    _TYPE   DATES   TYPE_CHANGES    DAYS_SINCE_LAST_EVENT
   1    12345   active  2021-01-15  0   1
   2    12345   active  2021-01-16  0   2
   3    12345   active  2021-01-17  0   3
   4    12345   dormant 2021-01-18  1   1
   5    12345   dormant 2021-01-19  1   2
   6    12345   churned 2021-01-20  2   1
   7    12345   churned 2021-01-21  2   2
   8    12345   churned 2021-01-22  2   3
   9    39498   active  2021-01-15  3   1
  10    39498   active  2021-01-16  3   2
  11    39498   dormant 2021-01-17  4   1
  12    39498   churned 2021-01-18  5   1

Answer 2

使用NVL、FIRST_VALUE、LAG、DATEDIFF、DATEADD、IFF

冗长的步骤如下所示：

SELECT 
   user,
   type,
   dates,
   lead(type)over(partition by user order by dates) as lead_event,
   iff(lead_event != type, dates, null) as diff_event_date,
   first_value(dates)over(partition by user order by dates) as first_user_date   
   lag(diff_event_date) ignore nulls over (partition by user order by dates) as last_event_date,
   NVL(last_event_date, dateadd(day,-1,first_user_date)) AS date_of_true_interest
   ​datediff(day, date_of_true_interest, dates) as "days since last event"
FROM data
ORDER BY 1,3

但这行不通。因此，如果我们需要将第二个滞后改组为第二个逻辑阶段。因此：

SELECT 
   user,
   type,
   dates,
   lag(diff_event_date) ignore nulls over (partition by user order by dates) as last_event_date,
   NVL(last_event_date, dateadd(day,-1,first_user_date)) AS date_of_true_interest,
   datediff(day, date_of_true_interest, dates) as "days since last event"
FROM (
    SELECT 
       user,
       type,
       dates,
       lead(type)over(partition by user order by dates) as lead_event,
       iff(lead_event != type, dates, null) as diff_event_date,
       first_value(dates)over(partition by user order by dates) as first_user_date   
    FROM data
)
ORDER BY 1,3;

这很丑陋，所以要整理另一层。

SELECT 
   user,
   type,
   dates,
   datediff(day, date_of_true_interest, dates) as "days since last event"
FROM (
    SELECT 
       user,
       type,
       dates,
       lag(diff_event_date) ignore nulls over (partition by user order by dates) as last_event_date,
       NVL(last_event_date, dateadd(day,-1,first_user_date)) AS date_of_true_interest
    FROM (
        SELECT 
           user,
           type,
           dates,
           lead(type)over(partition by user order by dates) as lead_event,
           iff(lead_event != type, dates, null) as diff_event_date,
           first_value(dates)over(partition by user order by dates) as first_user_date
        FROM data
    ) 
)
ORDER BY 1,3;

并与一些混合data

WITH data AS (
    select * from values 
        (12345,'active',to_date('1/15/21', 'MM/DD/YY')),
        (12345,'active',to_date('1/16/21', 'MM/DD/YY')),
        (12345,'active',to_date('1/17/21', 'MM/DD/YY')),
        (12345,'dormant',to_date('1/18/21', 'MM/DD/YY')),
        (12345,'dormant',to_date('1/19/21', 'MM/DD/YY')),
        (12345,'churned',to_date('1/20/21', 'MM/DD/YY')),
        (12345,'churned',to_date('1/21/21', 'MM/DD/YY')),
        (12345,'churned',to_date('1/22/21', 'MM/DD/YY')),
        (39498,'active',to_date('1/15/21', 'MM/DD/YY')),
        (39498,'active',to_date('1/16/21', 'MM/DD/YY')),
        (39498,'dormant',to_date('1/17/21', 'MM/DD/YY')),
        (39498,'churned',to_date('1/18/21', 'MM/DD/YY'))
    v( user, type, dates)
)

我们得到结果：

用户	类型	日期	自上次活动以来的天数
12345	积极的	2021-01-15	1
12345	积极的	2021-01-16	2
12345	积极的	2021-01-17	3
12345	休眠	2021-01-18	1
12345	休眠	2021-01-19	2
12345	搅动	2021-01-20	1
12345	搅动	2021-01-21	2
12345	搅动	2021-01-22	3
39498	积极的	2021-01-15	1
39498	积极的	2021-01-16	2
39498	休眠	2021-01-17	1
39498	搅动	2021-01-18	1