Group parents with same children - sql

Question

EDIT: This is way harder to explain that I though, constatly editing based on comments. Thank you all for taking interest.
I have a table like this
ID Type ParentID
1 ChildTypeA 1
2 ChildTypeB 1
3 ChildTypeC 1
4 ChildTypeD 1
5 ChildTypeA 2
6 ChildTypeB 2
7 ChildTypeC 2
8 ChildTypeA 3
9 ChildTypeB 3
10 ChildTypeC 3
11 ChildTypeD 3
12 ChildTypeA 4
13 ChildTypeB 4
14 ChildTypeC 4
and I want to group parents that have same children - meaning same number of children of same type.
From parent point of view, there is a finite set of possible configurations (max 10).
If any parent has same set of children (by ChildType), I want to group them together (in what I call a configuration).
ChildTypeA-D = ConfigA
ChildTypeA-C = ConfigB
ChildTypeA, B, E, F = ConfigX
etc.
The output I need is parents grouped by Configurations.
Config Group ParentID
ConfigA 1
ConfigA 3
ConfigB 2
ConfigB 4
I have no idea where to even begin.

Answer 1

I named your table t. Please try if this is what you are looking for.
It's show matched and unmatched.
It's looking for parentids with the same number of rows (t1.cnt = t2.cnt) and that all the rows are matched (having COUNT(*) = t1.cnt).
You can try it here
;with t1 as (select parentid, type, id, count(*) over (partition by parentid order by parentid) cnt from t),
t3 as
(
select t1.parentid parentid1, t2.parentid parentid2, count(*) cn, t1.cnt cnt1, t2.cnt cnt2, ROW_NUMBER () over (order by t1.parentid) rn
from t1 join t1 as t2 on t1.type = t2.type and t1.parentid <> t2.parentid and t1.cnt = t2.cnt
group by t1.parentid, t2.parentid, t1.cnt, t2.cnt
having COUNT(*) = t1.cnt
),
notFound as (
select t1.parentid, ROW_NUMBER() over(order by t1.parentid) rn
from t1
where not exists (select 1 from t3 where t1.parentid = t3.parentid1)
group by t1.parentid
)
select 'Config'+char((select min(rn)+64 from t3 as t4 where t3.parentid1 in (t4.parentid1 , t4.parentid2))) config, t3.parentid1
from t3
union all
select 'Config'+char((select max(rn)+64+notFound.rn from t3)) config, notFound.parentid
from notFound
OUTPUT
config parentid1
ConfigA 1
ConfigA 3
ConfigB 2
ConfigB 4
If id 14 was ChildTypeZ then parentid 2 and 4 wouldn't match. This would be the output:
config parentid1
ConfigA 1
ConfigA 3
ConfigC 2
ConfigD 4

Answer 2

I have happen to have similar task. The data I'm working with is a bit bigger scale so I had to find an effective approach to this. Basically I've found 2 working approaches.
One is pure SQL - here's a core query. Basically it gives you smallest ParentID with same collection of children, which you can then use as a group id (you can also enumerate it with row_number). As a small note - I'm using cte here, but in real world I'd suggest to put grouped parents into temporary table and add indexes on the table as well.
;with cte_parents as (
-- You can also use different statistics to narrow the search
select
[ParentID],
count(*) as cnt,
min([Type]) as min_Type,
max([Type]) as max_Type
from Table1
group by
[ParentID]
)
select
h1.ParentID,
k.ParentID as GroupID
from cte_parents as h1
outer apply (
select top 1
h2.[ParentID]
from cte_parents as h2
where
h2.cnt = h1.cnt and
h2.min_Type = h1.min_Type and
h2.max_Type = h1.max_Type and
not exists (
select *
from (select tt.[Type] from Table1 as tt where tt.[ParentID] = h2.[ParentID]) as tt1
full join (select tt.[Type] from Table1 as tt where tt.[ParentID] = h1.[ParentID]) as tt2 on
tt2.[Type] = tt1.[Type]
where
tt1.[Type] is null or tt2.[Type] is null
)
order by
h2.[ParentID]
) as k
ParentID GroupID
----------- --------------
1 1
2 2
3 1
4 2
Another one is a bit trickier and you have to be careful when using it. But surprisingly, it works not so bad. The idea is to concatenate children into big string and then group by these strings. You can use any available concatenation method (xml trick or clr if you have SQL Server 2017). The important part is that you have to use ordered concatenation so every string will represent your group precisely. I have created a special CLR function (dbo.f_ConcatAsc) for this.
;with cte1 as (
select
ParentID,
dbo.f_ConcatAsc([Type], ',') as group_data
from Table1
group by
ParentID
), cte2 as (
select
dbo.f_ConcatAsc(ParentID, ',') as parent_data,
group_data,
row_number() over(order by group_data) as rn
from cte1
group by
group_data
)
select
cast(p.value as int) as ParentID,
c.rn as GroupID,
c.group_data
from cte2 as c
cross apply string_split(c.parent_data, ',') as p
ParentID GroupID group_data
----------- -------------------- --------------------------------------------------
2 1 ChildTypeA,ChildTypeB,ChildTypeC
4 1 ChildTypeA,ChildTypeB,ChildTypeC
1 2 ChildTypeA,ChildTypeB,ChildTypeC,ChildTypeD
3 2 ChildTypeA,ChildTypeB,ChildTypeC,ChildTypeD