Finding max() within a max() or highest value in 1row-1col w. multi-value

Question

Hi,
I am trying to get a result set that has following criteria
- for rows using the same WEL_FK column, I'd like the latest date
and within the latest date the highest weight. (Sort of a max() within a max()) for each WEL_FK).
-if there is a row that has no other matching WEL_FK;s (like row 8)
it should still return that row.
PER_PK WEL_FK A_DATE WEIGHT
1 1 2010-10-22 4
2 1 2010-11-23 4
3 1 2010-10-22 6
4 3 2010-10-22 2
5 3 2010-11-23 2
6 3 2010-10-22 2
7 4 2010-10-22 5
8 6 2010-10-22 5
9 6 2010-10-21 6
10 6 2010-10-22 4
10 rows selected.
RESULTS desired
PER_PK WEL_FK A_DATE WEIGHT
2 1 2010-11-23 4 because it has the latest date within same foreign key even if wt. not the highest
5 3 2010-11-23 2 because it has the latest date within same foreign key, wts. are equal
7 4 2010-10-22 5 because there are no other rows with that WEL_FK
8 6 2010-10-22 5 because it has latest date with highest wt.
I am able to get close to what I want using the SQL below which gives me correct rows
but does not tell me the "weight" nor "PER_PK" column.
I have been trying to find a way to pick highest value from 1 ROW and within 1 column
with multiple-values but think I have dug myself into a hole.
When I add in the PER_PK above, it gives me of course all 10 rows again.
SELECT wel_fk,
LTRIM(MAX(SYS_CONNECT_BY_PATH(a_date,','))
KEEP (DENSE_RANK LAST ORDER BY curr),',') AS employees
FROM (SELECT wel_fk,
a_date,
ROW_NUMBER() OVER (PARTITION BY wel_fk ORDER BY a_date) AS curr,
ROW_NUMBER() OVER (PARTITION BY wel_fk ORDER BY a_date) -1 AS prev
FROM test)
GROUP BY wel_fk
CONNECT BY prev = PRIOR curr AND wel_fk = PRIOR wel_fk
START WITH curr = 1;
WEL_FK EMPLOYEES
1 2010-10-22,2010-10-22,2010-11-23
3 2010-10-22,2010-10-22,2010-11-23
4 2010-10-22
6 2010-10-21,2010-10-22,2010-10-22
==========================================================
I have been exploring analytics, connect by's, subqueries, wm_concat, SYS_CONNECT_BY_PATH ,etc. along with max()for days and get real close
but somehow it's off.
If you are able to provide actual SQL solution, ideally using what I have already, would be appreciated.
Thanks.
the_sql

Answer 1

Try this:
WITH src AS (
SELECT per_pk,
       wel_fk,
       a_date,
       weight,
       MAX( a_date ) OVER (PARTITION BY wel_fk ORDER BY a_date
                  RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING ) latest_date,
       last_value( weight ) OVER (PARTITION BY wel_fk ORDER BY a_date, weight 
                  RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING ) max_weight
FROM TEST
SELECT per_pk,
       wel_fk,
       a_date,
       weight
FROM src
WHERE a_date = latest_date
     AND weight = max_weight
PER_PK                 WEL_FK                 A_DATE                    WEIGHT                
2                      1                      2010/11/23                4                     
5                      3                      2010/11/23                2                     
7                      4                      2010/10/22                5                     
8                      6                      2010/10/22                5