Is it possible to rank with group condition?

Question

Is it possible to rank with group condition? for example I have following dataset.
RANK() OVER (partition BY PC.POLICY_NUMBER, PC.TRANSACTION_TYPE, COV_CHG_EFF_DATE order by PC.POLICY_NUMBER, COV_CHG_EFF_DATE, PC.TIMESTAMP_ENTERED) AS RNK,
POLICY_NUMBER    TRANSACTION_TYPE  COV_CHG_EFF_DATE  TIMESTAMP_ENTERED                        Rank
10531075PQ                           01           01/FEB/2009              15/SEP/2009 01:16:09.356663 AM       1
10531075PQ                           01           01/FEB/2009              15/SEP/2009 01:16:09.387784 AM       2
10531075PQ                           02           15/OCT/2009             16/OCT/2009 04:40:24.564928 PM       1
10531075PQ                           02           15/OCT/2009             16/OCT/2009 04:40:24.678118 PM       2
10531075PQ                           10           15/OCT/2009             16/OCT/2009 04:45:20.290117 PM       1
10531075PQ                           10           15/OCT/2009             16/OCT/2009 04:40:29.088737 PM       2
10531075PQ                           09           15/OCT/2009             16/OCT/2009 04:40:29.088737 PM       1 (expected 3)
10531075PQ                           06           17/OCT/2009             17/OCT/2009 04:45:20.290117 PM       1
10531075PQ                           07           17/OCT/2009             17/OCT/2009 04:40:29.088737 PM       1 (expected 2) I want to group rank based by transaction IDs. For ex, '09' and '10' as one set and '06' an '07' as another set. Instead of rank from start, I want rank continue for any occurance of '09' or '10'. In the above example, for the following row, I am expecting rank 3 as there are 2 transaction '10' already exist for same COV_CHG_EFF_DATE.
10531075PQ 09 15/OCT/2009 16/OCT/2009 04:40:29.088737 PM 1 (expected 3)
I wonder if it possible with Rank or another other analytical function. I am not looking for exact working code, but will appreciate if someone give me idea/tips. Sample table, and test data if anyone want experiment
drop table PC_COVKEY_PD;
create table PC_COVKEY_PD (
POLICY_NUMBER varchar(30),
TERM_IDENT varchar(3),
COVERAGE_NUMBER varchar(3),
TRANSACTION_TYPE varchar(3),
COV_CHG_EFF_DATE date,
TIMESTAMP_ENTERED timestamp
delete from PC_COVKEY_PD;
commit;
insert into PC_COVKEY_PD values ('10531075PQ', '021', '002', '01', to_date('01/FEB/2009','DD/MM/YYYY'), cast('15/SEP/2009 01:16:09.356663 AM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '001', '01', to_date('01/FEB/2009','DD/MM/YYYY'), cast('15/SEP/2009 01:16:09.387784 AM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '004', '02', to_date('15/OCT/2009','DD/MM/YYYY'), cast('16/OCT/2009 04:40:24.164928 PM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '004', '02', to_date('15/OCT/2009','DD/MM/YYYY'), cast('16/OCT/2009 04:40:24.264928 PM' as timestamp));
insert into PC_COVKEY_PD values ( '10531075PQ', '021', '005', '10', to_date('15/OCT/2009','DD/MM/YYYY'), cast('16/OCT/2009 04:40:24.364928 PM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '002', '10', to_date('15/OCT/2009','DD/MM/YYYY'), cast('16/OCT/2009 04:40:24.464928 PM' as timestamp));
insert into PC_COVKEY_PD values ( '10531075PQ', '021', '004', '09', to_date('15/OCT/2009','DD/MM/YYYY'), cast('16/OCT/2009 04:40:24.564928 PM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '004', '06', to_date('22/NOV/2011','DD/MM/YYYY'), cast('17/OCT/2009 04:40:24.564928 PM' as timestamp));
insert into PC_COVKEY_PD values ('10531075PQ', '021', '004', '07', to_date('22/NOV/2011','DD/MM/YYYY'), cast('17/OCT/2009 04:40:24.664928 PM' as timestamp));
commit;
SELECT POLICY_NUMBER,
       TERM_IDENT,
       COVERAGE_NUMBER,
       TRANSACTION_TYPE,
       COV_CHG_EFF_DATE,
       TIMESTAMP_ENTERED,
        RANK() OVER (partition BY PC.POLICY_NUMBER, PC.TERM_IDENT, PC.TRANSACTION_TYPE, PC.COV_CHG_EFF_DATE
                order by PC.POLICY_NUMBER, PC.TERM_IDENT, PC.COV_CHG_EFF_DATE, PC.TIMESTAMP_ENTERED) AS RNK
FROM PC_COVKEY_PD PC
ORDER BY PC.POLICY_NUMBER, PC.TERM_IDENT, PC.COV_CHG_EFF_DATE, PC.TIMESTAMP_ENTERED ;Edited by: 966820 on 30-Oct-2012 19:26

Answer 1

Hi,
966820 wrote:
I want to group rank based by transaction IDs. For ex, '09' and '10' as one set and '06' an '07' as another set. So, whenever you see '09', you want to treat it as '10', and whenever you see '06', you want to act as if it were '07'.
You can use CASE to map '09' to '10', and '06' to '07'. like this:
,       RANK () OVER ( PARTITION BY  policy_number
                              ,             term_ident
                ,            CASE  transaction_type
                                WHEN  '06'  THEN  '07'
                                WHEN  '09'  THEN  '10'
                                        ELSE  transaction_type
                           END
                ,            cov_chg_eff_date
                ORDER BY      timestamp_entered
                 )     AS rnkWhen you "PARTITION BY x", there's no point in saying "ORDER BY x" in the same analytic clause.
Thanks for posting the CREATE TABLE and INSERT statements.
Thanks, too, for posting your query, though, as mentioned already, it's unreadable. When posting any formatted text (and code should always be formatted), type these 6 characters:
\(small letters only, inside curly brackets) before and after each section of formatted text, to preserve spacing.
See the forum FAQ {message:id=9360002}