在这种情况下如何在不手动列出所有类别的情况下转换 table (BigQuery)？

Question

我有一个 table 架构大致如下：

user_id | segment_id | day

segment_id的取值范围不小：从1到70。对于 day，它是 0-2。

理想情况下，我想将此 table 转换为

user_id | segment_1_day_1_count | segment_2_day_1_count | ... segment_70_day_1_count | ... | segment_70_day_3_count | ... segment_1_count | segment_2_count | segment_3_count | day_1_count | day_2_count | day_3_count | total_count

粗略地说，对于每个 user_id 我想获得所有细分组合的计数：

按细分
按天
按段和天
总计

这看起来很像旋转，但我不确定是否有一种基于多列旋转的方法。

我目前的尝试如下：

SELECT
  user_id,
  segment_id,
  day,
  COUNT(*) OVER (PARTITION BY user_id, segment_id),
  COUNT(*) OVER (PARTITION BY user_id, day),
  COUNT(*) OVER (PARTITION BY user_id, segment_id, day),
  COUNT(*) OVER (PARTITION BY user_id)
FROM some_table

这是我需要的数据，但不是我想要的格式。

Answer 1

考虑以下方法

execute immediate (select '''select * from your_table 
pivot (count(*) for 'segment_' || segment_id || '_day_' || day || '_count' in (''' || 
string_agg('"segment_' || segment_id || '_day_' || day || '_count"', ',' order by day, segment_id) || '))'
from (select distinct segment_id from your_table),
(select distinct day from your_table));

如果应用于类似于您所描述的虚拟数据 - 输出为

Answer 2

is it possible to also add counts for segment, day and total separately?

当然可以。只是[相对]简单扩展我最初提出的解决方案

execute immediate (select '''
  select * from (
    select * 
    from your_table 
    pivot (count(*) for 'segment_' || segment_id || '_day_' || day || '_count' in (''' || list1 || '''))
  ) 
  join (
    select * 
    from (select * except(day) from your_table) 
    pivot (count(*) for 'segment_' || segment_id || '_count' in (''' || list2 || '''))
  )
  using(user_id)
  join (
    select * 
    from (select * except(segment_id) from your_table) 
    pivot (count(*) for 'day_' || day || '_count' in (''' || list3 || '''))
  )
  using(user_id)
  join (
    select user_id, count(*) total 
    from your_table
    group by user_id
  )
  using(user_id)
'''
from (
  select string_agg('"segment_' || segment_id || '_day_' || day || '_count"', ',' order by day, segment_id) list1
  from (select distinct segment_id from your_table), (select distinct day from your_table)
),(
  select string_agg('"segment_' || segment_id || '_count"', ',' order by segment_id) list2
  from (select distinct segment_id from your_table)
),(
  select string_agg('"day_' || day || '_count"', ',' order by day) list3
  from (select distinct day from your_table)
) 
)

如果应用于类似于您描述的虚拟数据 - 输出是

在这种情况下如何在不手动列出所有类别的情况下转换 table (BigQuery)？

How to pivot the table (BigQuery) in this case without listing all categories manually?

sql

pivot

google-bigquery