Groups having non-overlapping intervals

Question

Hi there,
I need help on some mind boggling overlap problem, or perhaps non-overlap problem.
Having the sample data below, I need to put together subcodes in groups having non-overlapping
intervals [MIN_VAL;MAX_VAL[
I have this table:
SQL>create table t (product   varchar2(1) not null
  2                 ,subcode   number(1)   not null, constraint subcode_chk check (subcode >= 0)
  3                 ,min_val   number(2)   not null, constraint min_val_chk check (min_val >= 0)
  4                 ,max_val   number(2)   not null, constraint max_val_chk check (max_val >= 0)
  5                 ,constraint t_pk primary key (product, subcode)
  6                 ,constraint t_val_chk check (min_val < max_val));
Table created.
SQL>
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 0, 0, 99);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 1, 0, 99);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 2, 18, 67);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 3, 20, 65);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 4, 0, 99);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 5, 45, 60);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 6, 20, 40);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 7, 0, 99);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 8, 60, 65);
1 row created.
SQL>insert into t (product, subcode, min_val, max_val) values ('A', 9, 0, 99);
1 row created.
SQL>commit
  2  /
Commit complete.
SQL>
SQL>  select product, subcode, min_val, max_val
  2      from t
  3     where product = 'A'
  4  order by product, subcode;
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL
A                0          0         99
A                1          0         99
A                2         18         67
A                3         20         65
A                4          0         99
A                5         45         60
A                6         20         40
A                7          0         99
A                8         60         65
A                9          0         99
10 rows selected.
SQL>The records of interest are subcodes 5,6,8 since they in certain cases can be considered as one subcode.
It is OK, that MAX_VAL of one record = MIN_VAL of other record. The main thing is that the subcodes
within a group are mutual disclosing on MIN_VAL, MAX_VAL.
SQL>  select product, subcode, min_val, max_val
  2      from t
  3     where product = 'A'
  4       and subcode in (5,6,8)
  5  order by min_val;
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL
A                6         20         40
A                5         45         60
A                8         60         65
SQL>I have started out by trying to solve it using lag/lead analytical functions, but without luck.
Next, I've come up with this hierarchial query, but I don't quite understand it myself, and that bothers me!
SQL>    select distinct -- This, distinct, bothers me!
  2                      product
  3                     ,subcode
  4                     ,min_val
  5                     ,max_val
  6  --                   ,connect_by_isleaf
  7  --                   ,connect_by_root subcode
  8        from t
  9       where connect_by_isleaf = 1 -- Why does this help me?
10  start with -- This, start with, seems "clumpsy"
11            min_val in (  select min_val
12                            from t
13                        group by product, subcode, min_val)
14  connect by nocycle -- This, nocycle, really bothers me!
15                     min_val > prior min_val
16                 and max_val <= prior max_val
17                 and product = prior product
18                 and subcode <> prior subcode
19    order by product
20            ,subcode
21            ,min_val
22            ,max_val;
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL
A                5         45         60
A                6         20         40
A                8         60         65
SQL>Currently I'm struggling with just identifying the three subcodes. In the perfect world this would be better output
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL   GROUP_FLAG
A                0          0         99
A                1          0         99
A                2         18         67
A                3         20         65
A                4          0         99
A                5         45         60            1
A                6         20         40            1
A                7          0         99
A                8         60         65            1
A                9          0         99Or even better, if using herarchial query:
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL   ROOT_SUBCODE
A                0          0         99              0
A                1          0         99              1
A                2         18         67              2
A                3         20         65              3
A                4          0         99              4
A                5         45         60              6
A                6         20         40              6
A                7          0         99              7
A                8         60         65              6
A                9          0         99              9Any help and inspiration would be much appreciated. But please don't get offended if I don't follow up the next 12-14 hrs.
Regards
Peter
BANNER
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64biEdited by: Peter on May 19, 2009 12:52 PM
- Changed line 15 in hierarchial query

Answer 1

Brillant as always, Frank. Ten points and mange tak for you.
(1) What is your concept of "group" in this problem? In what sense do subcodes 5, 6 and 8 form a group? Is it that they share the quality of not overlapping with some other row with the same product? That is, will there be at most two groups per product: rows that overlap with every other row, and rows that don't?By group I mean, that when isolated from other subcodes, a group does not overlap. For product A, I have 10 different subcodes. In certain situations I'm allowed to consider 5,6 and 8 as one, since they don't overlap. In general the data presented results in 8 groups, 7 groups having 1 subcode, 1 group having 3 subcodes.
(2) What you mean by "mutually disclosing"? Is it that the min_val to max_val ranges do not overlap?Yes. They are however allowed to be equal.
As to your query. Seems you're right, it might actually be that "simple". I changed slightly, allowing >= and <=
SQL> SELECT m.*,
  2         CASE
  3            WHEN
  4            EXISTS (
  5                     SELECT  NULL
  6                     FROM    t
  7                     WHERE   product = m.product
  8                     AND     (       min_val >= m.max_val
  9                             OR      max_val <= m.min_val
10                             )
11                   )
12         THEN 1 END group_flag
13    FROM t m;
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL GROUP_FLAG
A                0          0         99
A                1          0         99
A                2         18         67
A                3         20         65
A                4          0         99
A                5         45         60          1
A                6         20         40          1
A                7          0         99
A                8         60         65          1
A                9          0         99
10 rows selected.
SQL>
This assumes that min_val <= max_val on each row.Your assumption is perfectly correct, as expessed in check constraint on table t.
>
The rest of this message concerns the CONNECT BY query you posted, in case you want to understand it better.
When I run the CONNECT BY query you posted, I get these results:
P    SUBCODE    MIN_VAL    MAX_VAL
A          0          0         99
A          1          0         99
A          4          0         99
A          5         45         60
A          6         20         40
A          7          0         99
A          8         60         65
A          9          0         99
Ouch, my bad. Somehow I posted the wrong query. I have edited line 15 from
  15                    min_val >= prior min_valinto
  15                    min_val > prior min_val
The START WITH clause includes all rows: you might as well not have a START WITH clause:
10  start with -- This, start with, seems "clumpsy"
11            min_val in (  select min_val
12                            from t
13                        group by product, subcode, min_val)
That's great, somehow I was misled to believe that wew should always have a START WITH.
Thank you for your remainding comments on hierarchial query. I'll definitely have to play around with it to fully understand. With the edit mentioned, and
without a START WITH it now looks as,
SQL>  select distinct -- This, distinct, bothers me!
  2                       product
  3                      ,subcode
  4                      ,min_val
  5                      ,max_val
  6         from t
  7        where connect_by_isleaf = 1 -- Why does this help me?
  8   connect by nocycle -- This, nocycle, really bothers me!
  9                      min_val > prior min_val
10                  and max_val <= prior max_val
11                  and product = prior product
12                  and subcode <> prior subcode  -- Not necessary
13     order by product
14             ,subcode
15             ,min_val
16             ,max_val;
PRODUCT    SUBCODE    MIN_VAL    MAX_VAL
A                5         45         60
A                6         20         40
A                8         60         65
3 rows selected.
SQL>One final question. In your solution with EXISTS, you use < and >. In my rewritten example I use <= and >=. It seems that it still works, even without a condition on subcode != m.subcode in the subquery.
Maybe I'm too tired by now, but why is that? - just don't get it
Regards
Peter