Group ranges in table with non-time-based start and end columns - sql

Question

I need to find the ranges in consecutive data points that may have gaps between them based on double precision start and end columns.
For simplicity let's call them startPoint and endPoint which track a position on a line. The difference between the endPoint and the startPoint would denominate a distance. Over this "distance" particular force/effect signal values are captured and based on the values a state is stored on the table. Each row has a unique id identifier.
Thus, the table looks like the following:
| id | startPoint | endPoint | state |
|----|------------|----------|----------|
| 1 | 0.0 | 5.8 | Active |
| 2 | 5.8 | 7.1 | Inactive |
| 3 | 7.5 | 10.2 | Inactive |
| 4 | 10.2 | 11.3 | Inactive |
| 5 | 11.6 | 12.1 | Active |
| 6 | 12.1 | 12.9 | Active |
I have struggled to come up with a query that works in PostgresSQL that yields the following result:
| startGap | endGap | state |
|------------|----------|----------|
| 0.0 | 5.8 | Active |
| 5.8 | 7.1 | Inactive |
| 7.5 | 11.3 | Inactive |
| 11.6 | 12.9 | Active |
Any help would be greatly appreciated.

Answer 1

Hmmm . . . You can identify where a group starts using lag() and then use cumulative sums:
select min(startPoint) as startPoint, max(endPoint) as endPoint, state
from (select t.*,
sum( (prev_endPoint is distinct from startPoint)::int) over () as grp
from (select t.*,
lag(endPoint) over (partition by state order by startPoint) as prev_endPoint
from t
) t
) t
group by state, grp;
To be honest, floating point numbers are rather dangerous, because two could look the same. The sum() defining grp is probably better written as:
sum( (abs(prev_endPoint - startPoint) > 0.001)::int) over () as grp
I would also suggest that you switch to fixed point representation (numeric) rather than floating point.