Using the Rank() function in 8i

Question

I have a table that holds multiple transactions and amounts for each person (each person uniquely identified by an id). For simplicity's sake, the let's say that the table only has four columns and looks like this:
PID NUMBER(8)
TID VARCHAR2(10)
AMT NUMBER(14,2)
CMPN VARCHAR2(10)
I need to be able to sum the various rows by person, and then calc the total sum(amt)of various percentages of people from the table, for example - What is the total/combined sum(amt) for the top 10% of the people, 'top' meaning people with the highest sum(amt)?
Of course I know how many rows I need to sum, because I know how many distinct people exist.
Unfortunately, I can't use RowId, because the order by is executed after the RowId limiting clause, not before, so the query would limit the rows, but not the right rows. Rank should evaluate what I need, and then I'd like to only select rows > a certain rank.
I've tried the following:
select cid, sum(amt),
rank() over (partition by cmpn order by sum(amt))
from tableA
group by cid
but get this error:
rank() over (partition by cmpn order by sum(amt))
ERROR at line 2:
ORA-00979: not a GROUP BY expression
CMPN is actually the same value for all the rows in this instance, but I added because I couldn't get the query to run even this far without partitioning by something.
I also couldn't use 'where RANK > NNN' - error on that too - invalid column was the error, as though I didn't already know that column didn't exist in my table...
ANY help or ideas at all would be greatly appreciated...

Answer 1

I have been struggling with the same thing. As far as I can
tell, the syntax simply is not powerful enough. Teradata and
IBM, for example, allow the user to append a 'Qualify' statement
to the Rank() command. 'Qualify' allows the query writer
to 'qualify' how many rows are to be returned, similarly to the
way you were trying to access a variable called RANK.
I did have some luck using subqueries, but I would much prefer
to do this all in one command. I am hoping that Ratio To Report
may help overcome this problem, since it apparently ties
together multiple analytic statements.