Classify records based on matching table - sql

Question

I have two tables: ITEMS and MATCHING_ITEMS, as below:
ITEMS:
|---------------------|------------------|
| ID | Name |
|---------------------|------------------|
| 1 | A |
| 2 | B |
| 3 | C |
| 4 | D |
| 5 | E |
| 6 | F |
| 7 | G |
|---------------------|------------------|
MATCHING_ITEMS:
|---------------------|------------------|
| ID_1 | ID_2 |
|---------------------|------------------|
| 1 | 2 |
| 1 | 3 |
| 2 | 3 |
| 4 | 5 |
| 4 | 6 |
| 5 | 6 |
|---------------------|------------------|
The MATCHING_ITEMS table defines items that match each other, and thus belong to the same group, i.e. items 1,2, and 3 match with each other and thus belong in a group, and the same for items 4,5, and 6. Item 7 does not have a match belong to any group.
I now need to add a 'Group' column on the ITEMS table which contains a unique integer for each group, so it would look as follows:
ITEMS:
|---------------------|------------------|------------------|
| ID | Name | Group |
|---------------------|------------------|------------------|
| 1 | A | 1 |
| 2 | B | 1 |
| 3 | C | 1 |
| 4 | D | 2 |
| 5 | E | 2 |
| 6 | F | 2 |
| 7 | G | NULL |
|---------------------|------------------|------------------|
So far I have been using a stored procedure to do this, looping over each line in the MATCHING_ITEMS table and updating the ITEMS table with a group value. The problem is that I eventually need to do this for a table containing millions of records, and the looping method is far too slow.
Is there a way that I can achieve this without using a loop?

Answer 1

If you have all pairs of matches in the matching table, then you can just use the minimum id to assign the group. For this:
select i.*,
(case when grp_id is not null
then dense_rank() over (order by grp_id)
end) as grouping
from items i left join
(select mi.id_1, least(mi.id1, min(mi.id2)) as grp_id
from matching_items mi
group by mi.id_1
) mi
on i.id = mi.id_1;
Note: This works only if all pairs are in the matching items table. Otherwise, you will need a recursive/hierarchical query to get all the pairs.

Answer 2

You could use min and max at first, then dense_rank to assign group numbers:
select id, name, dense_rank() over (order by mn, mx) grp
from (
select distinct id, name,
min(id_1) over (partition by name) mn,
max(id_2) over (partition by name) mx
from items left join matching_items on id in (id_1, id_2))
order by id
demo

Answer 3

The pairs 2,3 and 5,6 in the Matching_items table seem redundant as they could be derived (if I am reading your question right)
Here is how I did it. I just reused id_1 from your example as the group no:
create table
items (
ID number,
name varchar2 (2)
);
insert into items values (1, 'A');
insert into items values (2, 'B');
insert into items values (3, 'C');
insert into items values (4, 'D');
insert into items values (5, 'E');
insert into items values (6, 'F');
insert into items values (7, 'G');
create table
matching_items (
ID number,
ID_2 number
);
insert into matching_items values (1, 2);
insert into matching_items values (1, 3);
insert into matching_items values (2, 3);
insert into matching_items values (4, 5);
insert into matching_items values (4, 6);
insert into matching_items values (5, 6);
with new_grp as
(
select id, id_2, id as group_no
from matching_items
where id in (select id from items)
and id not in (select id_2 from matching_items)),
assign_grp as
(
select id, group_no
from new_grp
union
select id_2, group_no
from new_grp)
select items.id, name, group_no
from items left outer join assign_grp
on items.id = assign_grp.id;