Need help with User Defined Aggregates - Statistical Mode

Question

I would like to use User Defined Aggregates (UDAG) to calculate the Statistical Mode. The mode is the most frequent value among a set of observations.
http://en.wikipedia.org/wiki/Mode_(statistics)
In my application I aggregate an address level table of 130 million records to a ZIP4 level table of roughly 36 million records. Some ZIP4s will have 1 record, some may have 1000+ records. I need to use statistical modes to pick the most frequent values for some of the attribute fields.
Presently I am using an approach from AskTom using the Analytic functions combined with nesting subqueries. The code works but it's cumbersome. I feel user defined aggregates should be able to perform a simpler more straightforward integration into SQL queries.
I've reviewed several of the other posts in this forum on User Defined Aggregates. I feel I could write a procedure that calculates and average or merely concatenates strings. But this particular application of the UDAGs is stumping me. Rather than just increased a running total or count or concatenating a master string, you'd have to keep every distinct values and a count for how many times that value occured. Then evaluate those items to pick the most frequent to pass to output. I'm finding it difficult as a novice.
Any, I'll post a quick example using the present approach that I want to replace with a UDAG:
Here's a small table:
DRD> desc statmodetest
Name Null? Type
ID NUMBER
STATE VARCHAR2(2)
REGION VARCHAR2(1)
1* select * from statmodetest
DRD> /
ID ST REG
1 TX W
2 MN W
3 CA W
4 VA E
5 VA E
6 KY E
7 MN W
8 FL E
9 OK W
10 NC E
11 TX W
12 WI E
13 CA W
14 MI E
15 FL E
16 FL E
17 TN E
18 FL E
19 WI E
20 MA E
Now here's a query approach that gets the MODE State value for each Region.
1 SELECT DISTINCT region, mode_state, mode_state_cnt FROM (
2 SELECT region,
3 FIRST_VALUE(state) OVER (PARTITION BY region ORDER BY stcnt DESC) AS mode_state,
4 FIRST_VALUE(stcnt) OVER (PARTITION BY region ORDER BY stcnt DESC) AS mode_state_cnt
5 FROM (
6 select id, state, region, COUNT(state) OVER (PARTITION BY region, state) AS stcnt
7* from statmodetest t ) )
DRD> /
R MO MODE_STATE_CNT
W CA 2
E FL 4
What I'd like to be able to do is have a UDAG that supports this style query:
SELECT region, UDAGMODE(state)
FROM statmodetest
GROUP BY region ;
Thanks,
Paul

Answer 1

This is not what you want..?
SQL> select * from test;
        ID STATU REGIO
         1 TX    W
         2 MN    W
         3 CA    W
         4 VA    E
         5 VA    E
         6 KY    E
         7 MN    W
         8 FL    E
         9 OK    W
        10 NC    E
        11 TX    W
        12 WI    E
        13 CA    W
        14 MI    E
        15 FL    E
        16 FL    E
        17 TN    E
        18 FL    E
        19 WI    E
        20 MA    E
20 rows selected.
SQL> select region,max(status) keep(dense_rank first order by cnt desc,status) st,
  2         max(cnt)
  3  from(
  4       select region,status,count(*) cnt
  5       from test
  6       group by region,status)
  7  group by region;
REGIO ST      MAX(CNT)
E     FL             4
W     CA             2
<br>
<br>
Or I misread..?