Select value of most occurances - analytical function?!

Question

Hi ...
I've get stuck with a "little" problem.
I try to provide some testing code for this:
CREATE TABLE a1 (
     id NUMBER(8),
     val NUMBER(6),
     title VARCHAR2(16),
     CONSTRAINT test_pk PRIMARY KEY(id)
INERT INTO a1 (id, val, title) VALUES (1,12,'c');
INERT INTO a1 (id, val, title) VALUES (2,13,'b');
INERT INTO a1 (id, val, title) VALUES (3,13,'a');
INERT INTO a1 (id, val, title) VALUES (4,13,'a');
INERT INTO a1 (id, val, title) VALUES (5,42,'a');
INERT INTO a1 (id, val, title) VALUES (6,42,'b');
INERT INTO a1 (id, val, title) VALUES (7,42,'b');Actually the table is much bigger ;) But this should be ok for this question. It already exist a query like:
SELECT
     count(*) -- just an example
FROM
     a1
GROUP BY
     val
-- should return 1,3,3 (for the groups val=12, val=13,val=42)Now it is nessecary to select a title for each group (specified by group by). And there the title which occurs the most in this group should be selected. For this example this are 'c' for the group val=12 .. 'a' for the group val=13 and finally 'b' for the group val=42.
I tried to use some anayltical function, but I#m not able to get this to work - may be because I never used analytical functions before. If I try something I mostly get an error: Keyword FROM not at expected position (ORA-00923). I searched for some tutorial/howto documentations where my problem is handled but without success. So I guess the syntax and the way to understand analytical functions is not as easy as it semms to be ...
title OVER ( ORDER BY count(*) ROWS | RANGE BETWEEN 1 AND 1 ) <-- that would by logical for my brain, but not for oracles ;-)
Can somebody help?
Thanks!

Answer 1

Hi folks,
thanks for the variuos answers! Weekend is over, so we should work on ...
I tried some examples you gave me. And I decide to provide more detailes information! At first, if the amount of 'a' and 'b' is equal it doesn't matter which one will be returned (so it can be undefined, if that makes thing easier). I will now paste the original query I work with and I add some comments for you - so you can find the lines which should be changed to return the most occurances-value.
If you think it makes sense to Provide some create-table ddl an (maybe as csv file) some data, tell me how i can do that (I think its not an option to post ~ Mio rows as Inserts here).
The select-query I want to manipulate - this is no more related to our test-table 'a1' !! For Example let us look on the rows where 'A' and 'drm_' is selected - starts at line 7!
SELECT
  box_id,
  schedule_id,
  fixsecs_down(MIN(acqtime),600),
  COUNT(*), -- each row in rfm_meas_hr represents one frame of measuremnt-data, so this represents the number of frames received in this block
  -- instead of 'A' the most occurance of the col 'rpro' should be selected here
  'A',
  -- like above, but from the column 'rdmo'
  'drm_',
  -- below this some other cols are calculated/selected, not important here
  FLOOR(MEDIAN(rfre)),
  ROUND(AVG(rdbv),2),
  ROUND(SUM(POWER(rdbv,2)),2),
  ROUND(MAX(rdbv),2),
  ROUND(MIN(rdbv),2),
  ROUND(SUM(rsnr)/SUM(nframes),2),
  ROUND(SUM(POWER(rsnr,2)),2),
  ROUND(MAX(rsnr),2),
  ROUND(MIN(rsnr),2),
  ROUND( AVG(rsta_sync),2), -- rsta_sync
  ROUND(SUM(POWER(rsta_sync,2)),2), -- rsta_sync_s
  ROUND( MIN(rsta_sync),2), -- rsta_sync_min
  ROUND( MAX(rsta_sync),2), -- rsta_sync_max
  ROUND( AVG(rsta_fac),2), -- rsta_facc
  ROUND(SUM(POWER(rsta_fac,2)),2), -- rsta_fac_s
  ROUND( MIN(rsta_fac),2), -- rsta_fac_min
  ROUND( MAX(rsta_fac),2), -- rsta_fac_max
  ROUND( AVG(rsta_sdc),2), -- rsta_sdc
  ROUND(SUM(POWER(rsta_sdc,2)),2), -- rsta_sdc_s
  ROUND( MIN(rsta_sdc),2), -- rsta_sdc_min
  ROUND( MAX(rsta_sdc),2), -- rsta_sdc_max
  ROUND( AVG(rsta_audio),2), -- rsta_audio
  ROUND(SUM(POWER(rsta_audio,2)),2), -- rsta_audio_s
  ROUND( MIN(rsta_audio),2), -- rsta_audio_min
  ROUND( MAX(rsta_audio),2), -- rsta_audio_max
  MIN(rser), TODO:  most occurances
  MIN(rtty_stream0), -- TODO:  most occurances
  MIN(rtty_stream1), -- TODO:  most occurances
  MIN(rtty_stream2), -- TODO:  most occurances
  MIN(rtty_stream3), -- TODO:  most occurances
  ROUND(AVG(NVL(rafs_error/nullif(rafs_au,0),1))*SUM(rafs_au)/NULLIF(SUM(rafs_au),0),2), -- rafs
  ROUND( SUM( POWER(  NVL(rafs_error/nullif(rafs_au,0),1),2))*SUM(rafs_au)/NULLIF(SUM(rafs_au),0) ,2), -- rafs_s
  ROUND(MIN(rafs_error/ NULLIF(rafs_au,0)),2), -- rafs_min
  ROUND(MAX(NVL(rafs_error/NULLIF(rafs_au,0),1) )*SUM(rafs_au)/NULLIF(SUM(rafs_au),0),2), -- rafs_max
  SUM(robm_A),
  SUM(robm_B),
  SUM(robm_C),
  SUM(robm_D),
  SUM(robm_E),
  ROUND(SUM(rwmf) / SUM(nframes),2), -- rwmf
  ROUND(SUM(POWER(rwmf,2)),2), -- rwmf_s
  ROUND(MIN(rwmf),2), -- rwmf_min
  ROUND(MAX(rwmf),2), -- rwmf_max
  ROUND(SUM(rwmm) / SUM(nframes),2), -- rwmm
  ROUND(SUM(POWER(rwmm,2)),2), -- rwmm_s
  ROUND(MIN(rwmm),2), -- rwmm_min
  ROUND(MAX(rwmm),2), -- rwmm_max
  ROUND(SUM(rmer) / SUM(nframes),2), -- rmer
  ROUND(SUM(POWER(rmer,2)),2), -- rmer_s
  ROUND(MIN(rmer),2), -- rmer_min
  ROUND(MAX(rmer),2), -- rmer_max
  ROUND(SUM(RBP0_ERRS+ RBP1_ERRS+ RBP2_ERRS+ RBP3_ERRS) / NULLIF(SUM(RBP0_BITS+ RBP1_BITS+ RBP2_BITS+ RBP3_BITS),0) ,10), -- ber
  ROUND(SUM(POWER( (RBP0_ERRS+ RBP1_ERRS+ RBP2_ERRS+ RBP3_ERRS) / NULLIF((RBP0_BITS+ RBP1_BITS+ RBP2_BITS+ RBP3_BITS),0) ,2)),10), -- ber_s
  ROUND(MIN(RBP0_ERRS+ RBP1_ERRS+ RBP2_ERRS+ RBP3_ERRS) / NULLIF(MIN(RBP0_BITS+ RBP1_BITS+ RBP2_BITS+ RBP3_BITS),0) ,10), -- ber_min
  ROUND(MAX(RBP0_ERRS+ RBP1_ERRS+ RBP2_ERRS+ RBP3_ERRS) / NULLIF(MAX(RBP0_BITS+ RBP1_BITS+ RBP2_BITS+ RBP3_BITS),0) ,10), -- ber_max
  ROUND(AVG(rdop),2), -- rdop
  ROUND(SUM(POWER(rdop,2) ),2), -- rdop_s
  ROUND(MIN(rdop),2), -- rdop_min
  ROUND(MAX(rdop),2), -- rdop_max
  ROUND(AVG(rdel90),2), -- rdel90
  ROUND(SUM(POWER(rdel90,2) ),2), -- rdel90_s
  ROUND(MIN(rdel90),2), -- rdel90_min
  ROUND(MAX(rdel90),2), -- rdel90_max
  ROUND(AVG(rdel95),2), -- rdel95
  ROUND(AVG(rdel99),2), -- rdel99
  null AS reslevel
FROM
-- select the data where this should be calculated
    SELECT
    FROM
      rfm_meas_hr
    WHERE
      acqtime < fixsecs_down(to_timestamp('07-01-2011 14:00:00,00','dd-mm-yyyy hh24:mi:ss,ff'),600)
      AND (reslevel IS NULL OR reslevel=10)
-- group the selected data and executes the calculation given by SELECT-statement
GROUP BY
  -- group the data into 10min packages, indicated by its timestamp
  to_char( EXTRACT(MONTH FROM acqtime)*100000 + EXTRACT(DAY FROM acqtime)*1000 + EXTRACT(HOUR FROM acqtime)*10 + floor(EXTRACT(MINUTE FROM acqtime)/10) ),
  schedule_id,
  box_id
HAVING
  SUM(nframes)>15
;I chould say: I can add indexes if nessecary! At the moment there is one on (acqtime, reslevel) as this improves the access speed. But the query above, executed on typical data already takes 5-7 sec.
Please let me know if you need any more information.
Regards!