Grouping Data 3 Hours after the Initial Time - sql

Question

I need to be able to filter down a dataset to only show the first instance every 3 hours. If an instance is found, any other instances that occur up to 3 hours afterwards should be hidden.
The closes thing I've been able to find is using date_trunc to get the first instance each hour, but I need to hide specifically up to 3 hours after the first instance exactly.
Example Data:
+------------------------+-------+
| Timestamp | Value |
+------------------------+-------+
| "2015-12-29 13:35:00" | 65 |
| "2015-12-29 13:40:00" | 26 |
| "2015-12-29 13:45:00" | 80 |
| "2015-12-29 13:50:00" | 10 |
| "2015-12-29 16:40:00" | 76 |
| "2015-12-29 16:45:00" | 73 |
| "2016-01-04 08:05:00" | 87 |
| "2016-01-04 08:10:00" | 90 |
| "2016-01-04 08:15:00" | 52 |
| "2016-01-04 08:20:00" | 90 |
| "2016-01-04 08:25:00" | 23 |
| "2016-01-04 08:30:00" | 96 |
| "2016-01-04 13:35:00" | 53 |
| "2016-01-04 13:40:00" | 15 |
| "2016-01-04 13:45:00" | 85 |
+------------------------+-------+
Expected Result:
+------------------------+-------+
| Timestamp | Value |
+------------------------+-------+
| "2015-12-29 13:35:00" | 65 |
| "2015-12-29 16:40:00" | 76 |
| "2016-01-04 08:05:00" | 87 |
| "2016-01-04 13:30:00" | 7 |
+------------------------+-------+
Anyone have any ideas? Thank you so much for your help.

Answer 1

This is tricky, because you need to keep track of the last picked record to identify the next one - so you can't just group by 3 hours intervals.
Here is one approach using a recursive cte:
with recursive cte(ts, value) as (
select ts, value
from mytable
where ts = (select min(ts) from mytable)
union all
select x.*
from (select ts from cte order by ts desc limit 1) c
cross join lateral (
select t.ts, t.value
from mytable t
where t.ts >= c.ts + interval '3' hour
order by t.ts
limit 1
) x
)
select * from cte order by ts
The idea is to start from the earliest record in the table, then iterate by picking the first available record that is at least 3 hours later (this assumes no duplicates in the timestamp column).
Note that timestamp is not a good choice for a column name, because it conflicts with a language keyword (that's a datatype). I remaned it to ts in the query.
Demo on DB Fiddle:
ts | value
:------------------ | ----:
2015-12-29 13:35:00 | 65
2015-12-29 16:40:00 | 76
2016-01-04 08:05:00 | 87
2016-01-04 13:35:00 | 53