SQL/BigQuery：如何避免对一个组的多个非连续成员进行分组？

Question

I'm encountering an issue I can't seem to solve myself. I am grouping rows by location and timestamp, and finding the first and last timestamps for instances where an entity remained stationary. The issue is that for my current code, SQL groups together rows when the entity returns to a location it has been before.

In my example, an entity is at location -66.89 10.5002 at 2020-05-24 05:22:00 and then returns to that location at 2020-05-24 11:13:00. The result of the current query makes it look like that entity was in that location for the entire time, although the rows in between clearly show it moved. This is a conceptual problem I really don't know how to solve in SQL. I'm doing this in Big Query but I remember hitting a similar wall in SQL Server.

Code:

with selection as (
select 1 as id,TIMESTAMP '2020-05-24 11:13:00' as timestamp_, 'POINT(-66.89 10.5002)' as geom
union all select
1,TIMESTAMP '2020-05-24 05:22:00','POINT(-66.89 10.5002)'
union all select
1,TIMESTAMP '2020-05-24 05:25:00','POINT(-66.8881 10.4994)'
union all select
1,TIMESTAMP '2020-05-24 09:14:00','POINT(-66.8888 10.4958)'
union all select
1,TIMESTAMP '2020-05-24 07:37:00 UTC','POINT(-66.8873 10.5)'
union all select
1, TIMESTAMP'2020-05-24 07:52:00 UTC','POINT(-66.8873 10.5)'
)

select id,timestamp_,geom,
first_value(timestamp_)
    OVER (PARTITION BY id,geom ORDER BY timestamp_ ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) AS interval_start,
last_value(timestamp_)
    OVER (PARTITION BY id,geom ORDER BY timestamp_ ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) AS interval_end,
FROM
selection order by id,timestamp_

Result. Note the interval_start and interval_end for the first and last row

id	timestamp_	geom	interval_start	interval_end
1	2020-05-24 05:22:00 UTC	POINT(-66.89 10.5002)	2020-05-24 05:22:00 UTC	2020-05-24 11:13:00 UTC
1	2020-05-24 05:25:00 UTC	POINT(-66.8881 10.4994)	2020-05-24 05:25:00 UTC	2020-05-24 05:25:00 UTC
1	2020-05-24 07:37:00 UTC	POINT(-66.8873 10.5)	2020-05-24 07:37:00 UTC	2020-05-24 07:52:00 UTC
1	2020-05-24 07:52:00 UTC	POINT(-66.8873 10.5)	2020-05-24 07:37:00 UTC	2020-05-24 07:52:00 UTC
1	2020-05-24 09:14:00 UTC	POINT(-66.8888 10.4958)	2020-05-24 09:14:00 UTC	2020-05-24 09:14:00 UTC
1	2020-05-24 11:13:00 UTC	POINT(-66.89 10.5002)	2020-05-24 05:22:00 UTC	2020-05-24 11:13:00 UTC

Desired result:

id	timestamp_	geom	interval_start	interval_end
1	2020-05-24 05:22:00 UTC	POINT(-66.89 10.5002)	2020-05-24 05:22:00 UTC	2020-05-24 05:22:00 UTC
1	2020-05-24 05:25:00 UTC	POINT(-66.8881 10.4994)	2020-05-24 05:25:00 UTC	2020-05-24 05:25:00 UTC
1	2020-05-24 07:37:00 UTC	POINT(-66.8873 10.5)	2020-05-24 07:37:00 UTC	2020-05-24 07:52:00 UTC
1	2020-05-24 07:52:00 UTC	POINT(-66.8873 10.5)	2020-05-24 07:37:00 UTC	2020-05-24 07:52:00 UTC
1	2020-05-24 09:14:00 UTC	POINT(-66.8888 10.4958)	2020-05-24 09:14:00 UTC	2020-05-24 09:14:00 UTC
1	2020-05-24 11:13:00 UTC	POINT(-66.89 10.5002)	2020-05-24 11:13:00 UTC	2020-05-24 11:13:00 UTC

Answer 1

Consider below

with selection as (
  select 1 as id,TIMESTAMP '2020-05-24 11:13:00' as timestamp_, 'POINT(-66.89 10.5002)' as geom union all select
  1,TIMESTAMP '2020-05-24 05:22:00','POINT(-66.89 10.5002)' union all select
  1,TIMESTAMP '2020-05-24 05:25:00','POINT(-66.8881 10.4994)' union all select
  1,TIMESTAMP '2020-05-24 09:14:00','POINT(-66.8888 10.4958)' union all select
  1,TIMESTAMP '2020-05-24 07:37:00 UTC','POINT(-66.8873 10.5)' union all select
  1, TIMESTAMP'2020-05-24 07:52:00 UTC','POINT(-66.8873 10.5)'
), pregrouped_selection as (
  select id, timestamp_, geom, 
    countif(flag) over(partition by id order by timestamp_) grp
  from (
    select id, timestamp_, geom,
      geom != ifnull(lag(geom) over(partition by id order by timestamp_), geom) flag,
    from selection 
  )
  order by id, timestamp_
)
select id,timestamp_,geom,
first_value(timestamp_)
    OVER (PARTITION BY id,grp ORDER BY timestamp_ ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) AS interval_start,
last_value(timestamp_)
    OVER (PARTITION BY id,grp ORDER BY timestamp_ ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) AS interval_end,
FROM
pregrouped_selection order by id,timestamp_    

with output

As you can see - I left your original query almost 100% as is - just replaced geom to grp inside over() statement AND from pregrouped_selection which does calculate group number - grp

Answer 2

You can check if there are at least two distinct values using window functions:

min(geom) over (partition by id) <> max(geom) over (partition by id) as has_moved,