Question

I’m looking for a query that takes mostly consecutive days and groups them together depending on whether they are really consecutive and on the value belonging to that date. I’m using Oracle version 11g.

This is some example data:

date          value
2012-01-01    2000
2012-01-02    2000 //(there is no data for Jan 03 for example)
2012-01-04    2000
2012-01-05    5000
2012-01-06    5000
2012-01-07    5000
2012-01-08    2000
2012-01-09    2000
2012-01-10    2000

(this is the result of quite a large query)

What I’m looking for would group those days together in periods like this:

from_date   to_date     value
2012-01-01  2012-01-02  2000
2012-01-04  2012-01-04  2000
2012-01-05  2012-01-07  5000
2012-01-08  2012-01-10  2000

We did manage to formulate a query that does what I want, but it’s not a very efficient way, and I’m pretty sure something better/more elegant exists. This is what I’m using now:

with temp_table as (
select a.pk_date DATE1, c.pk_date DATE2, a.volume VOL1
from dm_2203 a, dm_2203 c
where a.volume = c.volume
  and a.pk_date <= c.pk_date
  and not exists (select 1 from dm_2203 b
                  where a.volume = b.volume
                    and a.pk_date = b.pk_date+1)
  and not exists (select 1 from dm_2203 d
                  where c.volume = d.volume
                    and c.pk_date = d.pk_date-1)  )
select * from temp_table y
where date2-date1+1 = (select count(*)
                       from dm_2203 z
                       where z.pk_date between y.date1 and y.date2
                         and y.vol1 = z.volume)
order by 1;

Does anyone have an idea on how to do this faster and without all the joins? Thanks!

Answer 1

I thin this should work and also be reasonable efficient (it should hit the table only once)

create table t ( d date, v number);

insert into t values (trunc(sysdate), 100);
insert into t values (trunc(sysdate+2), 100);
insert into t values (trunc(sysdate+3), 100);
insert into t values (trunc(sysdate+4), 100);
insert into t values (trunc(sysdate+5), 200);
insert into t values (trunc(sysdate+6), 200);
insert into t values (trunc(sysdate+7), 200);
insert into t values (trunc(sysdate+8), 100);

select min(d), max(d), v
from (
    select d, v, 
        sum( gc) over (partition by v order by d) g
    from (
        select d, v, 
            (case (d - lag(d) over ( partition by v order by d) )
                when 1 then 0
                else 1
             end) gc
        from t
    )
) group by v, g
order by min(d), v

Note that if you want to run the logic in an efficient way on a subset of your data, you should add the where clause in the inner most select. Otherwise oracle will have problems to use any index.