Grouping rows based on a consecutive flag in SQL (Redshift) - sql

Question

I've got a tricky problem that I am trying to solve here and can't get my head around it so far.
So the problem is this: I have tracking data, where there are records produced over time. Let's say you have a robot driving around and you record it's position once every second. Each of those positions is recorded as one record in the database (we use AWS Redshift).
Each record has a tracking_id which is unique across all records that belong to the same source of the tracking, i.e. unique for the robot. Then I have a record_id which is globally unique, a timestamp, and a flag that indicates if the record was created while the robot was inside or outside a defined zone. And then there is some additional data like coordinates.
Here is a little illustration. The pink box is the zone, the green line is the path of the robot and the blue dots are the produced records.
So now I would like to group records based on the zone flag (have a look at the screenshot below). So I want to isolate sub-paths inside the zone into a record and grab the start and end timestamp and position. The IDs don't matter so I don't necessarily need to keep the tracking or record ids even though I listed them in the desired result.
Thanks for the help, I would really appreciate it! Also just solving part of the problem like how to group based on the flag without grabbing first and last values within the sub-paths would help already.

Answer 1

This is a gaps and islands problem. In this case, you want the islands where in_zone happens to be TRUE (and there are two of them). We can use the difference in row number method here:
WITH cte AS (
SELECT *, ROW_NUMBER() OVER (PARTITION BY tracking_id ORDER BY timestamp) rn1,
ROW_NUMBER() OVER (PARTITION BY tracking_id, in_zone ORDER BY timestamp) rn2
FROM yourTable
)
SELECT
tracking_id,
MIN(record_id) AS record_id,
MIN(timestamp) AS start_timestamp,
MAX(timestamp) AS end_timestamp,
(SELECT t2.coordinates FROM yourTable t2
WHERE t2.record_id = MIN(t1.record_id) AND t2.tracking_id = t1.tracking_id) AS entry_coordinates,
(SELECT t2.coordinates FROM yourTable t2
WHERE t2.record_id = MAX(t1.record_id) AND t2.tracking_id = t1.tracking_id) AS exit_coordinates
FROM cte t1
WHERE
in_zone = 'TRUE'
GROUP BY
tracking_id,
rn1 - rn2,
in_zone
ORDER BY
tracking_id,
record_id DESC;
Demo

Answer 2

This is a gaps-and-islands problem. I would approach it using LAG() to identify the previous in-group and a cumulative sum. You can also use conditional aggregation to get the first and last coordinate values:
SELECT tracking_id, MIN(record_id), MIN(timestamp) as start_timestamp,
MIN(timestamp) as end_timestamp,
MAX(CASE WHEN prev_in_zone IS NULL OR prev_in_zone <> in_zone THEN coordinates END) as entry_coordinates,
MAX(CASE WHEN next_in_zone IS NULL OR next_in_zone <> in_zone THEN coordinates END) as entry_coordinates
FROM (SELECT t.*,
SUM( CASE WHEN prev_in_zone = in_zone THEN 0 ELSE 1 END) OVER (PARTITION BY tracking_id ORDER BY timestamp ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) as grp
FROM (SELECT t.*,
LAG(in_zone) OVER (PARTITION BY tracking_id ORDER BY timestamp) as prev_in_zone,
LEAD(in_zone) OVER (PARTITION BY tracking_id ORDER BY timestamp) as next_in_zone
FROM t
) t
) t
WHERE in_zone = 'TRUE'
GROUP BY tracking_id, grp;
With much appreciation to Tim, here is a db<>fiddle.