top N (aggregate) for each group by

I am a little confused about your request, can you post a little more information.. Top 2 by number of occurences or something else?

I'm making a report from a log table.. here's a better example

time, deploymentid, domain, actionid
2/2/2004 1:23am, 1, aol.com, 1
2/2/2004 4:37am, 2, aol.com, 1
2/2/2004 5:11am, 3, aol.com, 2
2/3/2004 12:03pm, 3, yahoo.com, 1
2/3/2004 2:43pm, 3, msn.com, 1
2/3/2004 2:50pm, 2, aol.com, 2

I like to give the option to group by time, deployment, and domain. for instance.. deployment by domain would be this query: (with some example aggregates I may use)

select
deploymentid,
domain,
sum(case actionid when in (1) then 1 else 0 end) a1
group by deploymentid,domain
order by a1 desc

the same report, grouped by domain and then deployment would be:
select
domain,
deploymentid,
sum(case actionid when in (1) then 1 else 0 end) a1
group by domain,deploymentid
order by a1 desc

in either case, the domain column contains upwards of 30,000 unique values.. so the report gets cluttered with tons of insignificant data. what the customer is really interested in is the top N domains (top depends on the aggregate).. in the example above the top 2 domains would be the domains with the most rows with where actionid=1

oops, that is:
sum(case when actionid in (1) then 1 else 0 end)

Paste the following in your query analyser with appropriate fieldnames and tablenames and execute:




select group1, group2, count(*) cnt
into temp2 from stuff
group by group1, group2
order by group1, group2,cnt  asc

select a.group1, max(a.cnt) cnt into temp3
from temp2 a, temp2 b
where a.cnt < (Select max(b.cnt) from temp2 b)
group by a.group1

UNION

select a.group1, max(a.cnt) cnt
from temp2 a
inner join temp2 b
on a.group1 = b.group1
group by a.group1


select a.group1,a.group2,a.cnt from temp2 a inner join temp3 b
on a.group1 = b.group1 and a.cnt = b.cnt

drop table temp2
drop table temp3

thanks for the reply iyerbhuvanac, your solution does return only the top 2, however I do need to limit to an arbitrary number of results per group. top 2 was just for my example.. the actual options are top 5, 20, 50, 100.

HERE is a proc which is going to help you. You just have to pass the parameter ie the top number of rows you want to retrieve.say,5 or 10 or 15 or 20,etc

create this in your database  and execute with the required params.

This is the base version. I shall try to optimize this and post it as and when time permits
 

create procedure Getmax @param int
AS
declare
@count int
begin
set @count = 1

select group1, group2, count(*) cnt
into temp2 from stuff
group by group1, group2
order by group1, group2,cnt  asc


while @count <= @param
begin

select a.group1, max(a.cnt) cnt into temp3
from temp2 a
inner join temp2 b
on a.group1 = b.group1
group by a.group1


select a.group1,a.group2,a.cnt cnt into temp5 from temp2 a inner join temp3 b
on a.group1 = b.group1 and a.cnt = b.cnt

select * from temp5

delete from temp2   where temp2.group1 in (select group1 from temp5)
and temp2.group2 in (select group2 from temp5)
and temp2.cnt in (select cnt from temp5)

set @count = @count + 1


drop table temp3
drop table temp5
end

drop table temp2

end

I appreciate your effort, but I really needed something less hackerish (read scalable)

I came up with this method using a temp table..

select g1,g2,c1,c2
group by g1,g2
into #temp

create clustered index on #temp (g1,c1 desc) -- important!
select g1,g2,c1,c2 from temp t
where g2 in (select top 10 g2 from #temp where g1=t.g1 order by c1 desc)

thanks for everyones help