Conditional case statement help.!

Question

Hello Forum Members,
I have a table with stores order along with order dates and other information. I have a requirement to write a query to get the nominations from that order's table according to the logic mentioned below.
Nom1 =  Initial order date and next order within 7 days from initial order date then its considered as R1
Nom2 = next activity within 7 seven on the same ord_id from its R1 date , this goes on as the order dates increases for same  or single ord_no
Final Rn  is Count (Rn)
Rn %-- Count(Rn)  / Count Distinct ( ord_no ) * 100
Example data output:
ORD_DATE
ORD_ID
1_of_x
2_of_x
3_of_x
4_of_x
5_of_x
6_of_x
Final RnX
Rn%
7/10/2013
10095V1
1
0
0
0
0
0
8/3/2013
10095V1
1
0
0
0
0
0
8/8/2013
10095V1
0
1
0
0
0
0
8/12/2013
10095V1
0
0
1
0
0
0
9/6/2013
10095V1
1
0
0
0
0
0
DDL:
create table xx_ord_noms
ord_no varchar(30),
ord_date date,
ord_id varchar2(30)
Sample Data:
INSERT INTO xx_ord_noms (ord_no, ord_date, ord_id) VALUES ('10091', TO_DATE('07/10/2013','MM/DD/YYYY'), '10091');
INSERT INTO xx_ord_noms (ord_no, ord_date, ord_id) VALUES ('10091', TO_DATE('08/03/2013','MM/DD/YYYY'), '10091');
INSERT INTO xx_ord_noms (ord_no, ord_date, ord_id) VALUES ('10091', TO_DATE('08/08/2013','MM/DD/YYYY'), '10091');
INSERT INTO xx_ord_noms (ord_no, ord_date, ord_id) VALUES ('10091', TO_DATE('08/12/2013','MM/DD/YYYY'), '10091');
INSERT INTO xx_ord_noms (ord_no, ord_date, ord_id) VALUES ('10091', TO_DATE('09/06/2013','MM/DD/YYYY'), '10091');
commit;
I was planning to doing something like this but I am not getting the logic for R1 .. R7
select ord_no,
       ord_date,
       ord_id,
       prev_dt,
       (to_date(ord_date) - to_date(prev_dt)) data_diff,
       case
         when prev_dt is null or
              (to_date(ord_date) - to_date(prev_dt)) < 7 then
          'R1'
         when (to_date(ord_date) - to_date(prev_dt)) > 7 THEN
          'R2'
         ELSE
          nULL
       END
  FROM (select a.*,
               (SELECT MAX(tmp2.ord_date)
                  FROM xx_ord_noms tmp2
                 WHERE a.ord_id = tmp2.ord_id
                   AND tmp2.ord_date < a.ord_date) prev_dt
          from xx_ord_noms a);
Could someone give me advises or suggestions on how we can get the expected output.
Thanks in advance.

Answer 1

LAG and LEAD are good when you want to look forward or back a fixed number of rows, but in this case you don't know how far back you'll need to go.
I've got a solution. it calculates the Rn as a number, not a column position, but you can transform that with CASE statements if you want. I did it in two main steps:
First I calculated the Rn by comparing the date of the current and previous row. To do that, I generated a row number using the row_number()  analytic function:
select ord_no,
       ord_date,
       ord_id,
       numrow
, Rn
from xx_ord_noms
model
  partition by (ord_no)
  dimension by (row_number() over (partition by ord_no order by ord_date)  numrow)
  measures (ord_date, 0 Rn, ord_id )
  rules update
    rn[1] = 1,
    rn[ANY] = CASE when trunc(ord_date[cv()]) - trunc(ord_date[cv() - 1])  < 7 then rn[cv()-1]+1 else 1 end
ORD_NO     ORD_DATE             ORD_ID         NUMROW         RN
10091      10-Jul-2013 00:00:00 10091               1          1
10091      03-Aug-2013 00:00:00 10091               2          1
10091      08-Aug-2013 00:00:00 10091               3          2
10091      12-Aug-2013 00:00:00 10091               4          3
10091      06-Sep-2013 00:00:00 10091               5          1
Then I needed to identify all the rows of each series. I defined firstDt as the date of the first item in the series:
firstDt[1] = ord_date[cv()]
firstDt[ANY] = CASE when rn[cv()]= 1 then ord_date[cv()] else firstDt[cv()-1] end
With that, the FinalRn is just an analytic max(rn) over (partition by firstDt)
I'm not clear on what R% is. Your descriptions are unclear and I think they're inconsistent.
select ord_no, ord_date, ord_id
, Rn, finalRn, 100*finalRn/totCnt PCT1, 100 * totRn/totCnt PCT2
, totRn
from xx_ord_noms
model
  partition by (ord_no)
  dimension by (row_number() over (partition by ord_no order by ord_date)  numrow)
  measures (ord_date, 0 Rn, ord_id, to_date(null) firstDt, 0 finalRn, 0 totRn, 0 totCnt )
  rules upsert all
    rn[1] = 1
  , rn[ANY] = CASE when trunc(ord_date[cv()]) - trunc(ord_date[cv() - 1])  < 7 then rn[cv()-1]+1 else 1 end
  , firstDt[1] = ord_date[cv()]
  , firstDt[ANY] = CASE when rn[cv()]= 1 then ord_date[cv()] else firstDt[cv()-1] end
  , finalRn[ANY] = max(rn) over (partition by firstDt)
  , totCnt[ANY] = count(*) over()
  , totRn[ANY] = count(case when rn = 1 then 1 else null end) over ()
ORD_NO     ORD_DATE             ORD_ID             RN    FINALRN       PCT1       PCT2      TOTRN
10091      10-Jul-2013 00:00:00 10091               1          1         20         60          3
10091      03-Aug-2013 00:00:00 10091               1          3         60         60          3
10091      08-Aug-2013 00:00:00 10091               2          3         60         60          3
10091      12-Aug-2013 00:00:00 10091               3          3         60         60          3
10091      06-Sep-2013 00:00:00 10091               1          1         20         60          3
Regards,
David