Group data series into variable width windows based on first event - sql

Question

I have computational task which can be reduced to the follow problem:
I have a large set of pairs of integers (key, val) which I want to group into windows. The first window starts with the first pair p ordered by key attribute and spans all the pairs where p[i].key belongs to [p[0].key; p[0].key + N), with some arbitrary integer N, positive and common to all windows.
The next window starts with the first pair ordered by key not included in the previous windows and again spans all the pairs from its key to key + N, and so on for the following windows.
The last step is to sum second attribute for each window and display it together with the first key of the window.
For example, given list of records with values:
key
val
1
3
2
7
5
1
6
4
7
1
10
3
13
5
and N=3, the windows would be:
{(1,3),(2,7)},
{(5,1),(6,4),(7,1)},
{(10,3)}
{(13,5)}
The final result:
key
sum_of_values
1
10
5
6
10
3
13
5
This is easy to program with a standard programming language but I have no clue how to solve this with SQL.

Answer 1

Note: If clickhouse doesn't support the RECURSIVE keyword, just remove that keyword from the expression.
Clickhouse seems to use non-standard syntax for the WITH clause. The below uses standard SQL. Adjust as needed.
Sorry. clickhouse may not support this approach. If not, we would need to find another method of walking through the data.
Standard SQL:
There are a few ways. Here's one approach. First assign row numbers to allow recursively stepping through the rows. We could use LEAD as well.
Assign a group (key value) to each row based on the current key and the last group/key value and whether they are within some distance (N = 3, in this case).
The last step is to just SUM these values per group start_key and to use the start_key value as the starting key in each group.
WITH RECURSIVE nrows (xkey, val, n) AS (
SELECT xkey, val, ROW_NUMBER() OVER (ORDER BY xkey) FROM test
)
, cte (xkey, val, n, start_key) AS (
SELECT xkey, val, n, xkey FROM nrows WHERE n = 1
UNION ALL
SELECT t1.xkey, t1.val, t1.n
, CASE WHEN t1.xkey <= t2.start_key + (3-1) THEN t2.start_key ELSE t1.xkey END
FROM nrows AS t1
JOIN cte AS t2
ON t2.n = t1.n-1
)
SELECT start_key
, SUM(val) AS sum_values
FROM cte
GROUP BY start_key
ORDER BY start_key
;
Result:
+-----------+------------+
| start_key | sum_values |
+-----------+------------+
| 1 | 10 |
| 5 | 6 |
| 10 | 3 |
| 13 | 5 |
+-----------+------------+