如何在 BigQuery 中填写缺失的日期？答案

【问题标题】：How to fill missing dates in BigQuery?如何在 BigQuery 中填写缺失的日期？
【发布时间】：2021-04-07 18:14:57
【问题描述】：

此问题与 How to fill missing dates and values in partitioned data? 有关，但由于该解决方案不适用于 BigQuery，因此我再次发布此问题。

我有以下假设表格：

name       date          val
-------------------------------
A          01/01/2020     1.5
A          01/03/2020     2
A          01/06/2020     5
B          01/02/2020     90
B          01/07/2020     10

我想填写空白之间的日期并从最近的下一个日期复制该值。另外，我想填写以下日期：1）回到预设的 MINDATE（假设是 2019 年 12 月 29 日）和 2）回到当前日期（假设是 01/09/2020） - 对于 2)，默认值为 1。

所以，输出将是：

name       date          val
-------------------------------
A          12/29/2019     1.5
A          12/30/2019     1.5
A          12/31/2019     1.5
A          01/01/2020     1.5   <- original
A          01/02/2020     2
A          01/03/2020     2     <- original
A          01/04/2020     5
A          01/05/2020     5
A          01/06/2020     5     <- original
A          01/07/2020     1
A          01/08/2020     1
A          01/09/2020     1
B          12/29/2019     90
B          12/30/2019     90
B          12/31/2019     90
B          01/01/2020     90
B          01/02/2020     90    <- original
B          01/03/2020     10
B          01/04/2020     10
B          01/05/2020     10
B          01/06/2020     10
B          01/07/2020     10    <- original
B          01/08/2020     1
B          01/09/2020     1

上述问题中接受的solution 在 BigQuery 中不起作用。

【问题讨论】：

标签： python date datetime google-bigquery

【解决方案1】：

这应该可以工作

with base as (

select 'A' as name,           '01/01/2020' as date,     1.5 as val  union all
select 'A' as name,           '01/03/2020' as date,     2 as val union all
select 'A' as name,           '01/06/2020' as date,     5 as val union all
select 'B' as name,           '01/02/2020' as date,     90 as val union all
select 'B' as name,           '01/07/2020' as date,     10 as val
),

missing_dates as (

select name,dates as date from 
UNNEST(GENERATE_DATE_ARRAY('2019-12-29', '2020-01-09', INTERVAL 1 DAY)) AS dates cross join (select distinct name from base)

), joined as (
select distinct missing_dates.name, missing_dates.date,val 
from  missing_dates 
left join base on missing_dates.name = base.name 
and  parse_date('%m/%d/%Y', base.date) = missing_dates.date

)

select * except(val), 
ifnull(first_value(val ignore nulls) over(partition by name order by date ROWS BETWEEN CURRENT ROW AND
    UNBOUNDED FOLLOWING),1) as va1 
from joined

【讨论】：