基于单个日期从多行创建日期范围
Creating a Date Range from Multiple Rows Based on a Single Date
我有一个用户 table,具有以下字段:User_ID、电子邮件、Used_date。
正如我们所见,用户可以在一段时间内切换到多个电子邮件。我想从 used_date 字段创建日期范围字段(Email_Start_Date 和 Email_End_Date)。他们将存储用户使用该电子邮件的时间段。
用户可能会切换回旧电子邮件。在这种情况下,同一封电子邮件将有两个日期范围。
我还想填补上一封电子邮件的最后一天和当前电子邮件的开始日期之间的空白。
例如,如果用户在 2020 年 8 月 28 日 - 2020 年 8 月 31 日期间使用了 someone@gmail.com。
此外,他于 2020 年 9 月 3 日转为 someone1@gmail.com。
然后在输出中 someone@gmail.com 的日期范围为 8/28/2020 - 9/2/2020。
这是一个缺口和孤岛的例子。但是我不知道如何实现这个。
谢谢大家!
下次,将您的数据粘贴为文本,这样我们就不必再输入...
你是这个意思吗?我更喜欢“无限日期”而不是最后一个 until-date 的 NULL 值 - 我更喜欢“会话 ID”而不是“岛屿标识符”,它们通常在点击流和物联网分析中被称为 .. .
WITH
indata(userid,email,used_dt) AS (
SELECT 1,'someone@gmail.com' , DATE '2020-08-28'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-29'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-30'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-31'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-03'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-05'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-07'
UNION ALL SELECT 1,'someone@gmail.com', DATE '2020-09-09'
UNION ALL SELECT 2,'bob@gmail.com' , DATE '2019-07-12'
UNION ALL SELECT 3,'alice@newmail.com' , DATE '2020-08-08'
)
,
with_change_counter AS (
SELECT
userid
, email
, used_dt AS used_from_dt
, CASE
WHEN LAG(email,1,'') OVER(
PARTITION BY userid ORDER BY used_dt
) <> email
THEN 1
ELSE 0
END AS counter
, LEAD(used_dt,1,'9999-12-31') OVER(
PARTITION BY userid ORDER BY used_dt
) AS used_until_dt
FROM indata
)
,with_sess_id AS (
SELECT
userid
, email
, used_from_dt
, used_until_dt
, SUM(counter) OVER(PARTITION BY userid ORDER BY used_from_dt) AS sessid
, counter
FROM with_change_counter
)
SELECT
userid
, MAX(email) AS email
, MIN(used_from_dt) AS email_start_date
, MAX(used_until_dt) AS email_end_date
FROM with_sess_id
GROUP BY
sessid
, userid
ORDER BY
userid
, sessid
, email
;
-- out userid | email | email_start_date | email_end_date
-- out --------+--------------------+------------------+----------------
-- out 1 | someone@gmail.com | 2020-08-28 | 2020-09-03
-- out 1 | someone1@gmail.com | 2020-09-03 | 2020-09-09
-- out 1 | someone@gmail.com | 2020-09-09 | 9999-12-31
-- out 2 | bob@gmail.com | 2019-07-12 | 9999-12-31
-- out 3 | alice@newmail.com | 2020-08-08 | 9999-12-31
我只建议行号和聚合的区别:
select user_id, email, min(used_date) as email_start_date,
lead(min(used_date)) over (partition by user_id order by min(used_date)) - interval '1 day' as email_end_date
from (select t.*,
row_number() over (partition by user_id order by used_date) as seqnum,
row_number() over (partition by user_id, email order by used_date) as seqnum_2
from t
) t
group by user_id, email, (seqnum - seqnum_2);
实际上,您也可以使用 lag()
并且不聚合来做到这一点:
select user_id, email, min(used_date) as email_start_date,
lead(used_date) over (partition by user_id order by used_date) - interval '1 day' as email_end_date
from (select t.*,
lag(email) over (partition by user_id order by used_date) as prev_email
from t
) t
where prev_email is null or prev_email <> email;
第二个很简单。它只保留电子邮件更改的行(或用户数据开始的行)。然后它使用 lead()
获取结束日期。
Here 是一个 db<>fiddle.
我有一个用户 table,具有以下字段:User_ID、电子邮件、Used_date。
正如我们所见,用户可以在一段时间内切换到多个电子邮件。我想从 used_date 字段创建日期范围字段(Email_Start_Date 和 Email_End_Date)。他们将存储用户使用该电子邮件的时间段。
用户可能会切换回旧电子邮件。在这种情况下,同一封电子邮件将有两个日期范围。
我还想填补上一封电子邮件的最后一天和当前电子邮件的开始日期之间的空白。
例如,如果用户在 2020 年 8 月 28 日 - 2020 年 8 月 31 日期间使用了 someone@gmail.com。
此外,他于 2020 年 9 月 3 日转为 someone1@gmail.com。
然后在输出中 someone@gmail.com 的日期范围为 8/28/2020 - 9/2/2020。
这是一个缺口和孤岛的例子。但是我不知道如何实现这个。
谢谢大家!
下次,将您的数据粘贴为文本,这样我们就不必再输入...
你是这个意思吗?我更喜欢“无限日期”而不是最后一个 until-date 的 NULL 值 - 我更喜欢“会话 ID”而不是“岛屿标识符”,它们通常在点击流和物联网分析中被称为 .. .
WITH
indata(userid,email,used_dt) AS (
SELECT 1,'someone@gmail.com' , DATE '2020-08-28'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-29'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-30'
UNION ALL SELECT 1,'someone@gmail.com' , DATE '2020-08-31'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-03'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-05'
UNION ALL SELECT 1,'someone1@gmail.com', DATE '2020-09-07'
UNION ALL SELECT 1,'someone@gmail.com', DATE '2020-09-09'
UNION ALL SELECT 2,'bob@gmail.com' , DATE '2019-07-12'
UNION ALL SELECT 3,'alice@newmail.com' , DATE '2020-08-08'
)
,
with_change_counter AS (
SELECT
userid
, email
, used_dt AS used_from_dt
, CASE
WHEN LAG(email,1,'') OVER(
PARTITION BY userid ORDER BY used_dt
) <> email
THEN 1
ELSE 0
END AS counter
, LEAD(used_dt,1,'9999-12-31') OVER(
PARTITION BY userid ORDER BY used_dt
) AS used_until_dt
FROM indata
)
,with_sess_id AS (
SELECT
userid
, email
, used_from_dt
, used_until_dt
, SUM(counter) OVER(PARTITION BY userid ORDER BY used_from_dt) AS sessid
, counter
FROM with_change_counter
)
SELECT
userid
, MAX(email) AS email
, MIN(used_from_dt) AS email_start_date
, MAX(used_until_dt) AS email_end_date
FROM with_sess_id
GROUP BY
sessid
, userid
ORDER BY
userid
, sessid
, email
;
-- out userid | email | email_start_date | email_end_date
-- out --------+--------------------+------------------+----------------
-- out 1 | someone@gmail.com | 2020-08-28 | 2020-09-03
-- out 1 | someone1@gmail.com | 2020-09-03 | 2020-09-09
-- out 1 | someone@gmail.com | 2020-09-09 | 9999-12-31
-- out 2 | bob@gmail.com | 2019-07-12 | 9999-12-31
-- out 3 | alice@newmail.com | 2020-08-08 | 9999-12-31
我只建议行号和聚合的区别:
select user_id, email, min(used_date) as email_start_date,
lead(min(used_date)) over (partition by user_id order by min(used_date)) - interval '1 day' as email_end_date
from (select t.*,
row_number() over (partition by user_id order by used_date) as seqnum,
row_number() over (partition by user_id, email order by used_date) as seqnum_2
from t
) t
group by user_id, email, (seqnum - seqnum_2);
实际上,您也可以使用 lag()
并且不聚合来做到这一点:
select user_id, email, min(used_date) as email_start_date,
lead(used_date) over (partition by user_id order by used_date) - interval '1 day' as email_end_date
from (select t.*,
lag(email) over (partition by user_id order by used_date) as prev_email
from t
) t
where prev_email is null or prev_email <> email;
第二个很简单。它只保留电子邮件更改的行(或用户数据开始的行)。然后它使用 lead()
获取结束日期。
Here 是一个 db<>fiddle.