11g equivalent of Excel LARGE() function? (nth largest in set)

Question

I've been spinning my wheels on this and figured I'd give the forums a try.
I need to find the 13th largest value over the previous 52 weeks. I can set the window size correctly using ROWS 52 PRECEDING AND CURRENT ROW ORDER BY WEEK, but then I need another sort for the nTH largest within that partition. NTH_VALUE is close but what I really need is NTH_LARGEST_VALUE. Excel has a LARGE() function to do this.
Any thoughts for achieving this? See the example below:
Column desired_output is what I'm looking to achieve. It is the 2nd largest value over the current row and the previous 2 rows (ordered by week). Assume no relationship between week and val. These values just used for illustration purposes.
with tab as
    select to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7 week,
           to_number(to_char(to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7,'dd')) val,
           case when level in (1,2) then null
                when level = 3 then 11
                when level = 4 then 18
                when level = 5 then 18
                when level = 6 then 8
                when level = 7 then 8
                when level = 8 then 15
                when level = 9 then 22
                when level = 10 then 22
           end desired_result
      from dual connect by level <= 10
select week, val, desired_result
  from tab
order by week; 
WEEK             VAL DESIRED_RESULT
04-JAN-12          4
11-JAN-12         11
18-JAN-12         18             11
25-JAN-12         25             18
01-FEB-12          1             18
08-FEB-12          8              8
15-FEB-12         15              8
22-FEB-12         22             15
29-FEB-12         29             22
07-MAR-12          7             22
10 rows selected.Any advice would be appreciated. Thanks!

Answer 1

Ok,
Maybe this one :SQL> !cat q.sql
with tab as
    select to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7 week,
           to_number(to_char(to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7,'dd')) val,
           case when level in (1,2) then null
                when level = 3 then 11
                when level = 4 then 18
                when level = 5 then 18
                when level = 6 then 8
                when level = 7 then 8
                when level = 8 then 15
                when level = 9 then 22
                when level = 10 then 22
           end desired_result
      from dual connect by level <= 10
------ end of sample data ------
select fweek,val,desired_result, lval as calculated_result
from (
     select fweek,val,desired_result,src,lval, row_number() over (partition by fweek order by lval desc nulls last) r
     from (
          select fweek,val,desired_result,lval,src
          from (
               select to_char(week,'yyyymmdd') fweek, val, desired_result
               , lag(val,0) over (order by week asc) lag00
               , lag(val,1) over (order by week asc) lag01
               , lag(val,2) over (order by week asc) lag02
               from tab t
          unpivot include nulls (
               lval for (src) in
                    (lag00 as 0
                    ,lag01 as 1
                    ,lag02 as 2
where r=2
SQL> @q
FWEEK           VAL DESIRED_RESULT CALCULATED_RESULT
20120104          4
20120111         11                                4
20120118         18             11                11
20120125         25             18                18
20120201          1             18                18
20120208          8              8                 8
20120215         15              8                 8
20120222         22             15                15
20120229         29             22                22
20120307          7             22                22
10 rows selected.I have a difference from your desired_result for 20120111 : To me the 2nd greatest value for the 2 previous week is 4. I don't get why you want a NULL instead.
In order to have a bigger range (6 in the following example, but can be completed up to 52 weeks), the code should look like that :SQL> !cat q2.sql
with tab as
    select to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7 week,
           to_number(to_char(to_date('4-jan-2012', 'dd-mon-yyyy')+(level-1)*7,'dd')) val,
           case when level in (1,2) then null
                when level = 3 then 11
                when level = 4 then 18
                when level = 5 then 18
                when level = 6 then 8
                when level = 7 then 8
                when level = 8 then 15
                when level = 9 then 22
                when level = 10 then 22
           end desired_result
      from dual connect by level <= 10
------ end of sample data ------
select fweek,val,desired_result, lval as calculated_result
from (
     select fweek,val,desired_result,src,lval, row_number() over (partition by fweek order by lval desc nulls last) r
     from (
          select fweek,val,desired_result,lval,src
          from (
               select to_char(week,'yyyymmdd') fweek, val, desired_result
               , lag(val,0) over (order by week asc) lag00
               , lag(val,1) over (order by week asc) lag01
               , lag(val,2) over (order by week asc) lag02
               , lag(val,3) over (order by week asc) lag03
               , lag(val,4) over (order by week asc) lag04
               , lag(val,5) over (order by week asc) lag05
               , lag(val,6) over (order by week asc) lag06
               -- this should continue up to "lag(val,52) over (order by week asc) lag52"
               from tab t
          unpivot include nulls (
               lval for (src) in
                    (lag00 as 0
                    ,lag01 as 1
                    ,lag02 as 2
                    ,lag03 as 3
                    ,lag04 as 4
                    ,lag05 as 5
                    ,lag06 as 6
                    -- this should continue up to "lag52 as 52"
where r=3
SQL> @q2
FWEEK           VAL DESIRED_RESULT CALCULATED_RESULT
20120104          4
20120111         11
20120118         18             11                 4
20120125         25             18                11
20120201          1             18                11
20120208          8              8                11
20120215         15              8                15
20120222         22             15                18
20120229         29             22                22
20120307          7             22                22
10 rows selected.This is the 3rd value over the 6 previous weeks.
<b><u>Nota :</u></b>
For some unknown reason, I had to convert week to char because of <i>"ORA-01801: date format is too long for internal buffer"</i>
I didn't try to change my NLS_DATE_FORMAT to see if that would help.