Group rows based on column sum value

问题

I have a table with 3 columns as shown below:

id   | num_rows                         id   | num_rows | group_id
-----|---------                         -----|----------|--------
2502 | 330                              2502 | 330      | 9
3972 | 150                              3972 | 150      | 9
3988 | 200          =============>      3988 | 200      | 8
4228 | 280          Desired output      4228 | 280      | 8
3971 | 510          =============>      3971 | 510      | 1
52   | 1990                             52   | 1990     | 2
895  | 2000                             895  | 2000     | 3
812  | 5596                             812  | 5596     | 4
1600 | 7462                             1600 | 7462     | 5
910  | 7526                             910  | 7526     | 6
638  | 11569                            638  | 11569    | 7

id is a unique identifier for something while num_rows correspond to the number of rows each id has in another table.

I would like to group the rows (i.e., id column) such that the sum of num_rows is never above a specified value (in this case lets say 500).

Simply put: I want to group the id's in buckets with no bucket have rows more than 500. If an id is bigger than the limit then it gets its own separate group/bucket.

So far, I have been able to separate out the larger id's using the following query but I am not able to create groups for the remaining subset of the id's.

SELECT id, 
        num_rows,
        SUM(CASE WHEN num_rows > 500 THEN 1 ELSE 0 END) OVER(ORDER BY num_rows) AS group_id
FROM myTable;

id   | num_rows | group_id
-----|----------|--------
2502 | 330      | 0
3972 | 150      | 0
3988 | 200      | 0
4228 | 280      | 0
3971 | 510      | 1
52   | 1990     | 2
895  | 2000     | 3
812  | 5596     | 4
1600 | 7462     | 5
910  | 7526     | 6
638  | 11569    | 7

Thank you.

回答1:

I personally would prefere a pl/sql function for this task, but if you want to do it in pure sql you can use the following query:

WITH ord AS (SELECT id, num_rows, ROWNUM ord FROM myTable)
   , rek(ord, id, num_rows, sum_rows, groupId) AS 
         (SELECT ord, id, num_rows, num_rows, 1 FROM ord WHERE ord = 1
          UNION ALL
          SELECT rek.ord +1
               , ord.id
               , ord.num_rows
               , CASE WHEN rek.sum_rows + ord.num_rows > 500
                      THEN ord.num_rows
                      ELSE rek.num_rows + ord.num_rows
                END
               , CASE WHEN rek.sum_rows + ord.num_rows > 500
                      THEN rek.groupID + 1
                      ELSE rek.groupID
                 END
            FROM rek
            JOIN ORD
              ON ord.ord = rek.ord+1)
SELECT id, num_rows, groupid
  FROM rek
/

Note that this query does not search for matching entries to build groups such that the sum is < 500 as this is closely related to the so called knapsack problem (s. https://en.wikipedia.org/wiki/Knapsack_problem), which is all but easy to solve...

回答2:

If you don't want it by sequence, you can just make group using no of rows as follows

SELECT id, 
        num_rows,
        ceil(num_rows/500) AS group_id
FROM myTable;

This should new id for each 500 rows blocks.

来源：https://stackoverflow.com/questions/52218969/group-rows-based-on-column-sum-value

标签

sql

Oracle

group-by

grouping

rows