Identify not most recent candidate

Question

First Im sorry for creating a lengthy post, just trying to make it clear.
Please help me to make my query efficient.
Example follwed with logic behind the scenes.
create table finalcand (ptsequence number,version number,candidate varchar2(10),date_submitted date,
file_valid varchar(2))
insert into finalcand values(101,1,'VJ',TRUNC(SYSDATE),'Y');
insert into finalcand values(102,1,'RD',TRUNC(SYSDATE),'Y');
insert into finalcand values(103,1,'GD,TRUNC(SYSDATE),'Y');
insert into finalcand values(104,1,'GD',TRUNC(SYSDATE),'N');
insert into finalcand values(105,2,'GD,TRUNC(SYSDATE),'Y');
insert into finalcand values(107,2,'GVR',TRUNC(SYSDATE),'N');
insert into finalcand values(106,1,'GD',TRUNC(SYSDATE+1),'Y');
insert into finalcand values(108,1,'GD',TRUNC(SYSDATE),'Y');
insert into finalcand values(109,2,'GVR',TRUNC(SYSDATE+1),'Y');required o/p
   PTSEQUENCE    VERSION    PATIENT DATE_SUBMITTED FILE_VALID   RNK
     105     2     GD     7/23/2009           Y             2
     108     1     GD     7/23/2009           Y             3
     103     1     GD     7/23/2009           Y             4
     104     1     GD     7/23/2009           N             4
     107     2     GVR     7/23/2009           N             2im using this query which will work fine ,but im still trying to see if i can do it in a efficient way.
SELECT z.*
  FROM (SELECT r.*,
               RANK() OVER(
               PARTITION BY r.candidate
               ORDER BY R.DATE_SUBMITTED DESC,
                        R.VERSION DESC,
                        R.PTSEQUENCE desc) rnk
          FROM finalcand r
          where file_valid='Y') z
WHERE RNK >1
union all
SELECT z.*
  FROM (SELECT r.*,
               RANK() OVER(
               PARTITION BY r.candidate 
               ORDER BY R.DATE_SUBMITTED DESC,
                        R.VERSION DESC,
                        R.PTSEQUENCE desc) rnk
          FROM finalcand r
          ) z
WHERE RNK >1
and z.file_valid='N'Logic behind this
Problem is to identify all records which are not most latest candidate
-->candidate is identified by candidate name.
-->version and date_submitted is used to identify latest entry of that candidate.
-->if same candidate has multiple entries with same version and also same date_submitted then use max(ptsequence) to identify latest.
Catch here is
-->If we see a candidate entry with File_valid ='N' that means earlier version of that candidate should be the most latest.
Another catch is
-->If we see a candidate entry with File_valid ='N' and doesn’t have any older records then the current one should be latest.
Thanks for reading this post.
Edited by: user5495111 on Jul 23, 2009 1:37 PM

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statemnts! That helps a lot.
Did they work when you ran them? A couple of of the INSERT statemnets seem to be missing quotes.
Sentinel and I have slightly different ideas of what you want.
It looks like the criteria are different for the different values of file_valid.
If you add this row to the sample data:
insert into finalcand values(901,1,'VJ',TRUNC(SYSDATE),'N');Sentinel's solution includes the new row; your original solution and the one below do not.
WITH     got_rnk         AS
     SELECT     finalcand.*
     ,     RANK () OVER ( PARTITION BY  candidate
                           ,          file_valid
                           ORDER BY          date_submitted     DESC
                      ,               version          DESC
                      ,          ptsequence          DESC
                      )  AS rnk_fv
     ,     RANK () OVER ( PARTITION BY  candidate
                           ORDER BY          date_submitted     DESC
                      ,               version          DESC
                      ,          ptsequence          DESC
                      )  AS rnk_all
     FROM     finalcand
SELECT     ptsequence
,     version
,     candidate
,     date_submitted
,     file_valid
,     CASE  file_valid
          WHEN  'N'  THEN  rnk_all
          WHEN  'Y'  THEN      rnk_fv
     END     AS rnk
FROM     got_rnk
WHERE     (     rnk_fv          > 1
     AND     file_valid     = 'Y'
OR     (     rnk_all          > 1
     AND     file_valid     = 'N'
;I suspect this will be more efficient than your original query, since it only goes through the table once.