GROUP BY help

Question

Hey all,
I am having a difficult time with this one, any help would be appreciated.
I have a table with 4 fields:
ID,MAIN,TIME,SPD
it would look something like this
ROWID,ID, MAIN, TIME, SPD
1, AA, 01, 1, 45
2, AA, 02, 1, 44
3, AA, 02, 2, 33
4, AA, 03, 3, 44
5, AA, 02, 5, 55
etc
What I need to do is get an AVERAGE of the SPD column
where I GROUP by ID, MAIN and (here is the hard part) TIME (ONLY if its sequential)
So by ROWID
1 would stay the same,
2 and 3 would be averaged because ID, MAIN are the same and TIME is sequential
4 and 5 would stay the same...
looked at the AVG analytical function to see if I could just create a moving average (partitioned by ID, MAIN) but it still ends up averaging rows 2,3 and 5. I imagine you could do it with PL/SQL but not very versed in cursors and the table I would be running this on numbers in the billions of rows...
Thanks ahead of time,
Eric

Answer 1

If i understand clearly.
with t as (
select 1 r1, 'AA' id , 01 m1, 1 t1 , 45 spd from dual union all
select 2, 'AA', 02, 1, 44 from dual union all
select 3, 'AA', 02, 2, 33 from dual union all
select 4, 'AA', 03, 3, 44 from dual union all
select 5, 'AA', 02, 5, 55 from dual )
select id,m1,avg(spd)
from (select m1,r1,id,t1,grp,spd
from t
model
partition by (m1)
dimension by (row_number() over (partition by m1 order by r1) rn )
measures (r1,id,t1,spd,lag(t1) over ( partition by m1 order by t1) next_t1,1 grp, 1 grprec)
rules
upsert
iterate(1000)
until(presentv(t1[iteration_number+2],1,0)=0)
(grprec[any] = nvl(grp[cv()-1],1),
grp[any]= case when nvl( next_t1[cv()]+1,t1[cv()]) =t1[cv()]  then               
             grprec[cv()] else  grprec[cv()]+1  end))
group by id,m1,grpoutput
AA     3     44
AA     2     55
AA     1     45
AA     2     38.5
Edited by: alvinder on Mar 4, 2009 9:51 AM
Fixed the mistake i did.