Row lumping, cycle dates

user_8875256 0
Viewed 50

I want to look at the lead type and if that type is the same for that row then merge in those dates to fit within one row.

I have the below table:

id    start_dt     end_dt    type
 1     1/1/19      2/21/19   cross
 1     2/22/19     6/5/19    cross
 1     6/6/19      8/31/19   cross
 1     9/1/19      10/3/19   AAAA
 1     10/4/19     10/4/19   cross
 1     10/5/19     10/6/19   AAAA
 1     10/7/19     10/10/19  AAAA
 1     10/11/19    12/31/99  cross

Expected Results:

 id    start_dt    end_dt    type
 1     1/1/19      8/31/19   cross
 1     9/1/19      10/3/19   AAAA
 1     10/4/19     10/4/19   cross
 1     10/5/19     10/10/19  AAAA
 1     10/11/19    12/31/99  cross

How can I get my output to look like the expected results?

I have tested withlead lag rank and case expression but nothing worthy of adding here. Am I on the right path?

gaps-and-islands cycle date oracle sql
3 Answers

This is a gaps-and-islands problem. One option for solving it through contribution of row_number() analytical function :

select min(start_dt) as startdate, max(end_dt) as enddate, type
  from
  (
   with t(id, start_dt, end_dt,type) as
   (
    select 1, date'2019-01-01', date'2019-02-21', 'cross' from dual union all
    select 1, date'2019-02-22', date'2019-06-05', 'cross'  from dual union all
    select 1, date'2019-06-06', date'2019-08-31', 'cross'  from dual union all  
    select 1, date'2019-09-01', date'2019-10-03', 'AAAA'  from dual union all
    select 1, date'2019-09-04', date'2019-10-04', 'cross'  from dual union all   
    select 1, date'2019-10-05', date'2019-10-06', 'AAAA'  from dual union all
    select 1, date'2019-10-07', date'2019-10-10', 'AAAA'  from dual union all  
    select 1, date'2019-10-11', date'2019-12-31', 'cross'  from dual 
   )
   select type, 
          row_number() over (partition by id, type order by end_dt) as rn1,
          row_number() over (partition by id order by end_dt) as rn2,
          start_dt, end_dt
     from t
   ) tt
 group by type, rn1 - rn2
 order by enddate;

STARTDATE   ENDDATE     TYPE
---------   ---------   -----
01-JAN-19   31-AUG-19   cross
01-SEP-19   03-OCT-19   AAAA
04-SEP-19   04-OCT-19   cross
05-OCT-19   10-OCT-19   AAAA
11-OCT-19   31-DEC-19   cross

Demo

user_58413060

I actually think this is a pretty good case for Oracle's Pattern Matching Functionality.

with t(id, start_dt, end_dt,type) as
   (
    select 1, date'2019-01-01', date'2019-02-21', 'cross' from dual union all
    select 1, date'2019-02-22', date'2019-06-05', 'cross'  from dual union all
    select 1, date'2019-06-06', date'2019-08-31', 'cross'  from dual union all  
    select 1, date'2019-09-01', date'2019-10-03', 'AAAA'  from dual union all
    select 1, date'2019-09-04', date'2019-10-04', 'cross'  from dual union all   
    select 1, date'2019-10-05', date'2019-10-06', 'AAAA'  from dual union all
    select 1, date'2019-10-07', date'2019-10-10', 'AAAA'  from dual union all  
    select 1, date'2019-10-11', date'2019-12-31', 'cross'  from dual 
   )
SELECT *
FROM t
MATCH_RECOGNIZE(ORDER BY start_dt
                MEASURES a.id AS ID,
                         A.start_dt AS START_DT,
                         NVL(LAST(B.end_dt), A.end_dt) AS END_DT,
                         a.type AS TYPE
                PATTERN (A B*)
                DEFINE B AS start_dt > PREV(start_dt) AND type = PREV(type));

A detailed primer on the topic can be found here

user_22997600

If you want to look at adjacent rows to find groups that can combine, then I recommend lag() to find where groups start and a cumulative sum on that:

select id, type, min(start_dt), max(end_dt)
from (select t.*,
             sum(case when prev_end_dt >= start_dt - 1 then 0 else 1 end) over (partition by id, type order by start_dt) as grp
      from (select t.*,
                   lag(end_dt) over (partition by id, type order by start_dt) as prev_end_dt
            from t
           ) t
     ) t
group by id, type, grp
order by id, min(start_dt);

In particular, this will find cases where the type does not change but there is a gap in the time frames, as shown by this db<>fiddle for id = 2.

user_11440350
Related