在Oracle DB中使用SQL的数据集中每个组的最小（从）最大（到）日期

Question

I have following initial dataset

F_ID L_CAT CHG_DT
F1   VHL   01-FEB-2016
F1   VHL   10-FEB-2016
F1   VHL   15-FEB-2016
F1   MHL   20-FEB-2016
F1   VHL   25-FEB-2016
F1   VHL   28-FEB-2016
F1   MHL   05-MAR-2016
F1   MHL   10-MAR-2016
F2   VHL   01-FEB-2016
F2   VHL   10-FEB-2016
F2   MHL   18-FEB-2016
F2   MHL   21-FEB-2016
F2   VHL   25-FEB-2016

and want to generate following output using SQL Query in oracle DB

F_ID L_CAT FROM_DT      TO_DT
F1   VHL   01-FEB-2016  20-FEB-2016
F1   MHL   20-FEB-2016  25-FEB-2016
F1   VHL   25-FEB-2016  05-MAR-2016
F1   MHL   05-MAR-2016  10-MAR-2016
F2   VHL   01-FEB-2016  18-FEB-2016
F2   MHL   18-FEB-2016  25-FEB-2016
F2   VHL   25-FEB-2016  25-FEB-2016

In other words, I want to calculate time span during which each F_ID remain in specific L_CAT. I am using Oracle 11g. Any lead towards the solution is highly appreciated. Thanks

Code to produce the scenario is given under:

create table my_test
(
f_id varchar2(30),
l_cat varchar2(30),
chg_dt date
);

insert into my_test(f_id, l_cat, chg_dt) values ('F1','VHL','01-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','VHL','10-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','VHL','15-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','MHL','20-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','VHL','25-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','VHL','28-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','MHL','05-MAR-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F1','MHL','10-MAR-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F2','VHL','01-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F2','VHL','10-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F2','MHL','18-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F2','MHL','21-FEB-2016');
insert into my_test(f_id, l_cat, chg_dt) values ('F2','VHL','25-FEB-2016');
COMMIT;

Answer 1

This specific problem is called "gaps and islands". One method uses the difference of row numbers:

select f_id, l_cat, min(chg_dt), max(chg_dt)
from (select i.*,
             row_number() over (partition by f_id order by chg_dt) as seqnum_i,
             row_number() over (partition by f_id, l_cat order by chg_dt) as seqnum_ic
      from initial i
     ) i
group by f_id, l_cat, (seqnum_i - seqnum_ic);

Explaining how this works is challenging. But if you stare at the results from the subquery, you can see how the difference in row numbers defines the groups you want.

Answer 2

First of all, thank you for test case you provided!

How about this?

SQL> with inter as
  2    (select f_id, l_cat, chg_dt,
  3       lead(chg_dt) over (partition by f_id order by chg_dt) lead_dt,
  4       case when lag(l_cat, 1, 1) over (order by f_id, chg_dt) <> l_cat
  5            then 1
  6       end sgrp                                      -- group rows per L_CAT changes
  7       from my_test
  8    ),
  9  inter_2 as
 10    (select f_id, l_cat, chg_dt, lead_dt,
 11       sum(sgrp) over (order by f_id, chg_dt) grp    -- groups
 12     from inter
 13    )
 14  select f_id, l_cat,
 15    min(chg_dt) from_dt,
 16    nvl(max(lead_dt), min(chg_dt)) to_dt
 17  from inter_2
 18  group by f_id, l_cat, grp
 19  order by 1, 3;

F_ID  L_CAT      FROM_DT     TO_DT
----- ---------- ----------- -----------
F1    VHL        01-feb-2016 20-feb-2016
F1    MHL        20-feb-2016 25-feb-2016
F1    VHL        25-feb-2016 05-mar-2016
F1    MHL        05-mar-2016 10-mar-2016
F2    VHL        01-feb-2016 18-feb-2016
F2    MHL        18-feb-2016 25-feb-2016
F2    VHL        25-feb-2016 25-feb-2016

7 rows selected.

SQL>