A/b test groups

Question

I’ve to do a A/B test by count certain members in one group vs not in another group.
Background, a customer can purchase multiple products and they will have a product mix.
Customer A purchased products 101, 102, 103, then the product mix is ^101^102^103^, where ^ is a separator.
Note: product mix is a sorted distinct list
I’ve it working in slow performing rate (when the mix gets to the thousands, it runs for very long time ~20hrs).
Is there a faster / cleaner method?
Now, I have 2 tables
1. MIX --- Counts by PRODUCT_MIX for all my customers
2. LKUP --- Distinct PRODUCT list
Sample Data:
MIX               CNT     
^101^102^  10
^102^103^  5
^103^           1
LKUP
101
102
103
TAB1
PRODUCT      CONTROL_CNT    TEST_CNT
101                  6                             10
102                  1                             15
103                 10                           6
CREATE TABLE TAB1
TABLESPACE XXX_TBS COMPRESS NOLOGGING
AS 
SELECT
      lkup.PRODUCT,
      --------------- (CONTROL GROUP) only return someone did not have the PRODUCT in their MIX
      SUM(CASE
        WHEN instr(src.PRODUCT_MIX, '^'||lkup.PRODUCT||'^') = 0 THEN src.CNT
        ELSE NULL
      end) as CONTROL_CNT,
      --------------- (TEST GROUP) only return when someone has the PRODUCT in their MIX
      sum(case
        when instr(src.PRODUCT_MIX, '^'||lkup.PRODUCT||'^') > 0 then src.CNT
        else null
      end) as TEST_CNT
FROM
  ----- We use LKUP as the driver table eventhough this is a cartesian operation
  ----- the key here is to get all possible combo, but base on if the PRODUCT matches it will
  ----- either put them into CONTROL or TEST group.
  PRODUCT_MIX SRC,
  PRODUCT_LKUP LKUP
GROUP BY lkup.PRODUCT;Thank You
Edited by: BluShadow on 30-Mar-2011 12:26
added {noformat}{noformat} tags                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           

Answer 1

not sure I totally understand but you could try this.
step 1 make a row for up to the count number for each row so
^101^102^ cnt = 10 gets 10 rows and ^102^103^ cnt = 5 gets 5 rows.
with t as
select mix, cnt, rn from  (select '^101^102^'  mix, 10 cnt from dual union
                                select '^102^103^', 5  cnt from dual union
                                select '^103^',  1 from dual
MODEL
partition by (cnt)
DIMENSION BY (0 d)
measures (mix, cnt-1 i,  0 rn)
RULES
      ITERATE (100)  until iteration_number = i[0]
      (mix [ITERATION_NUMBER] = mix[0],
       rn[ITERATION_NUMBER] = iteration_number + 1)
select mix  cnt,  rn
from t
order by cnt, rn
CNT     RN
^101^102^     1
^101^102^     2
^101^102^     3
^101^102^     4
^101^102^     5
^101^102^     6
^101^102^     7
^101^102^     8
^101^102^     9
^101^102^     10
^102^103^     1
^102^103^     2
^102^103^     3
^102^103^     4
^102^103^     5
^103^     1step 2 break up the deliminiter of ^ into new rows so ^102^103^ would become two rows a 102 and a 103.
with t as
select mix, cnt, rn from  (select '^101^102^'  mix, 10 cnt from dual union
                                select '^102^103^', 5  cnt from dual union
                                select '^103^',  1 from dual
MODEL
partition by (cnt)
DIMENSION BY (0 d)
measures (mix, cnt-1 i,  0 rn)
RULES
      ITERATE (100)  until iteration_number = i[0]
      (mix [ITERATION_NUMBER] = mix[0],
       rn[ITERATION_NUMBER] = iteration_number + 1)
select mix --, cnt,  rn
from t
MODEL RETURN UPDATED ROWS
partition by (cnt, rn)
dimension by (0 d)
measures (mix)
RULES
ITERATE (1000) UNTIL mix[ITERATION_NUMBER] IS NULL
(mix [ITERATION_NUMBER + 1] =
     REGEXP_SUBSTR (mix[0],
                                     '[^\^]+',
                                   1,
                              ITERATION_NUMBER + 1)
order by mix
MIX
101
101
101
101
101
101
101
101
101
101
102
102
102
102
102
102
102
102
102
102
102
102
102
102
102
103
103
103
103
103
103now using this table it should be easy to count up the number of each group.
you said when the mix gets to the thousands
in that case you  might want to increase ITERATE (1000) to a higher number.Edited by: pollywog on Mar 30, 2011 8:33 AM