Combine three consecutive dates

Question

I am trying to combine three contiguous dates. I have web audit log table and if a user goes to the website on three consecutive days, then I want to record that as one visit not three. For example, if a user visits the website on 1 May, 2 May, 3 May, 5 May, 7 May, 8 May, 9 May, then it should be a total of 3 visits. One visit for the first three days, second visit for 5 May, and third visit for 7, 8 and 9 of May.
Is there any analytic function which can perform this kind of requirement.
All suggestions will be much appreciated. Thanks.

Answer 1

Papin
Awesome, man. You rock. Thanks a million for the
elegant code.First, I tried myself using analytic functions to write this query.
The following query is what I got.
It's quite similar to Papin Aziat's solution.
with 4 inline views.
SQL> with
  2    auditlog_tst as
  3      (
  4        select 0 as alog_mbr_no, to_date('01012007','ddmmyyyy') as visit_date from dual union all
  5        select 0 as alog_mbr_no, to_date('02012007','ddmmyyyy') as visit_date from dual union all
  6        select 0 as alog_mbr_no, to_date('03012007','ddmmyyyy') as visit_date from dual union all
  7        select 0 as alog_mbr_no, to_date('04012007','ddmmyyyy') as visit_date from dual union all
  8        select 0 as alog_mbr_no, to_date('10012007','ddmmyyyy') as visit_date from dual union all
  9        select 0 as alog_mbr_no, to_date('11012007','ddmmyyyy') as visit_date from dual union all
10        select 0 as alog_mbr_no, to_date('12012007','ddmmyyyy') as visit_date from dual union all
11        select 0 as alog_mbr_no, to_date('15012007','ddmmyyyy') as visit_date from dual
12      )
13   select alog_mbr_no, visit_date
14     from (select ttt.*,
15                  lead(m) over(partition by alog_mbr_no, rn order by visit_date) lead_m
16             from (select tt.*,
17                          sum(s) over(partition by alog_mbr_no, rn order by visit_date) + 1 ss,
18                          mod(sum(s) over(partition by alog_mbr_no,
19                                   rn order by visit_date) + 1,
20                              3) m
21                     from (select t.*,
22                                  decode(visit_date - 1,
23                                         lag(visit_date) over(partition by alog_mbr_no order by visit_date),
24                                         1,
25                                         0) s,
26                                  visit_date - row_number() over(partition by alog_mbr_no order by visit_date) rn
27                             from auditlog_tst t) tt) ttt)
28    where m = 1 or (m = 2 and lead_m is null)
29  /
ALOG_MBR_NO VISIT_DATE
          0 01.01.2007
          0 04.01.2007
          0 10.01.2007
          0 15.01.2007
SQL> but then I tried to solve it using model clause.
That's what I got:
SQL> with
  2      auditlog_tst as
  3        (
  4          select 0 as alog_mbr_no, to_date('01012007','ddmmyyyy') as visit_date from dual union all
  5          select 0 as alog_mbr_no, to_date('02012007','ddmmyyyy') as visit_date from dual union all
  6          select 0 as alog_mbr_no, to_date('03012007','ddmmyyyy') as visit_date from dual union all
  7          select 0 as alog_mbr_no, to_date('04012007','ddmmyyyy') as visit_date from dual union all
  8          select 0 as alog_mbr_no, to_date('10012007','ddmmyyyy') as visit_date from dual union all
  9          select 0 as alog_mbr_no, to_date('11012007','ddmmyyyy') as visit_date from dual union all
10          select 0 as alog_mbr_no, to_date('12012007','ddmmyyyy') as visit_date from dual union all
11          select 0 as alog_mbr_no, to_date('15012007','ddmmyyyy') as visit_date from dual
12        )
13    select mbr, dt from
14     (select * from auditlog_tst
15       model
16        partition by (alog_mbr_no mbr)
17        dimension by (row_number() over (partition by alog_mbr_no order by visit_date) rn)
18        measures (visit_date dt)
19         rules (dt[any] order by rn = case when dt[CV()]-2 = dt[CV()-2] and dt[CV()-1] is not null
20                                           then null
21                                           else dt[CV()]
22                                      end,
23                dt[ANY] order by rn = case when dt[CV()+1] is present and dt[CV()+1] is null
24                                           then null
25                                           else dt[CV()]
26                                      end)
27     ) where dt is not null
28  /
       MBR DT
         0 01.01.2007
         0 04.01.2007
         0 10.01.2007
         0 15.01.2007
SQL> Looks simpler, isn't it?
Let's compare three solutions:
SQL> drop table auditlog_tst;
Table dropped
SQL> create table auditlog_tst (alog_mbr_no number(5), visit_date date);
Table created
SQL> set timing on;
SQL> select * from auditlog_tst;
ALOG_MBR_NO VISIT_DATE
Executed in 0,109 seconds
SQL> declare dt date:=trunc(sysdate);
  2  begin
  3   for i in 1..100000
  4   loop
  5    dt:=dt+trunc(2*dbms_random.value)+1;
  6    insert into auditlog_tst values (trunc(i/1000),dt);
  7   end loop;
  8  end;
  9  /
PL/SQL procedure successfully completed
Executed in 22,766 seconds
SQL> select count(1) from auditlog_tst;
  COUNT(1)
    100000
Executed in 0,11 seconds
SQL> exec dbms_stats.gather_table_stats(user,'auditlog_tst');
PL/SQL procedure successfully completed
Executed in 2,719 seconds
SQL> delete from plan_table;
0 rows deleted
Executed in 0,172 seconds
SQL> explain plan for select alog_mbr_no,
  2         to_char(dt, 'dd/MON/YYYY', 'NLS_DATE_LANGUAGE = AMERICAN') visit_date
  3    from (select v1.*,
  4                 count(1) over(partition by alog_mbr_no, grp, xyz) cnt,
  5                 case
  6                   when count(1) over(partition by alog_mbr_no, grp, xyz) = 3 then
  7                    min(visit_date) over(partition by alog_mbr_no, grp, xyz)
  8                   else
  9                    visit_date
10                 end dt
11            from (select alog_mbr_no,
12                         visit_date,
13                         grp,
14                         trunc((row_number()
15                                over(partition by alog_mbr_no,
16                                     grp order by visit_date) - 1) / 3) xyz
17                    from (select t.*,
18                                 visit_date + 1 - row_number() over(order by alog_mbr_no, visit_date) as grp
19                            from auditlog_tst t) v0) v1) v2
20   group by alog_mbr_no, dt
21  /
Explained
Executed in 0,031 seconds
SQL> select * from table(dbms_xplan.display);
PLAN_TABLE_OUTPUT
Plan hash value: 2441421398
| Id  | Operation                | Name         | Rows  | Bytes |TempSpc| Cost (
|   0 | SELECT STATEMENT         |              |   101 |  1212 |       |  2351
|   1 |  HASH GROUP BY           |              |   101 |  1212 |       |  2351
|   2 |   VIEW                   |              |   100K|  1171K|       |  2337
|   3 |    WINDOW SORT           |              |   100K|  4003K|    10M|  2337
|   4 |     VIEW                 |              |   100K|  4003K|       |  1260
|   5 |      WINDOW SORT         |              |   100K|  2734K|  7864K|  1260
|   6 |       VIEW               |              |   100K|  2734K|       |   468
|   7 |        WINDOW SORT       |              |   100K|   976K|  3944K|   468
|   8 |         TABLE ACCESS FULL| AUDITLOG_TST |   100K|   976K|       |    53
15 rows selected
Executed in 0,141 seconds
SQL>  explain plan for select alog_mbr_no, visit_date
  2     from (select ttt.*,
  3                  lead(m) over(partition by alog_mbr_no, rn order by visit_date) lead_m
  4             from (select tt.*,
  5                          sum(s) over(partition by alog_mbr_no, rn order by visit_date) + 1 ss,
  6                          mod(sum(s) over(partition by alog_mbr_no,
  7                                   rn order by visit_date) + 1,
  8                              3) m
  9                     from (select t.*,
10                                  decode(visit_date - 1,
11                                         lag(visit_date)
12                                         over(partition by alog_mbr_no order by
13                                              visit_date),
14                                         1,
15                                         0) s,
16                                  visit_date - row_number() over(partition by alog_mbr_no order by visit_date) rn
17                             from auditlog_tst t) tt) ttt)
18    where m = 1
19       or (m = 2 and lead_m is null)
20  /
Explained
Executed in 0,032 seconds
SQL> select * from table(dbms_xplan.display);
PLAN_TABLE_OUTPUT
Plan hash value: 774399910
| Id  | Operation               | Name         | Rows  | Bytes |TempSpc| Cost (%
|   0 | SELECT STATEMENT        |              |   100K|  4687K|       |  1335
|*  1 |  VIEW                   |              |   100K|  4687K|       |  1335
|   2 |   WINDOW BUFFER         |              |   100K|  4003K|       |  1335
|   3 |    VIEW                 |              |   100K|  4003K|       |  1335
|   4 |     WINDOW SORT         |              |   100K|  3027K|  8664K|  1335
|   5 |      VIEW               |              |   100K|  3027K|       |   468
|   6 |       WINDOW SORT       |              |   100K|   976K|  3944K|   468
|   7 |        TABLE ACCESS FULL| AUDITLOG_TST |   100K|   976K|       |    53
Predicate Information (identified by operation id):
   1 - filter("M"=1 OR "M"=2 AND "LEAD_M" IS NULL)
19 rows selected
Executed in 0,156 seconds
SQL> explain plan for   select mbr, dt from
  2     (select * from auditlog_tst
  3       model
  4        partition by (alog_mbr_no mbr)
  5        dimension by (row_number() over (partition by alog_mbr_no order by visit_date) rn)
  6        measures (visit_date dt)
  7         rules (dt[any] order by rn = case when dt[CV()]-2 = dt[CV()-2] and dt[CV()-1] is not null
  8                                           then null
  9                                           else dt[CV()]
10                                      end,
11                dt[ANY] order by rn = case when dt[CV()+1] is present and dt[CV()+1] is null
12                                           then null
13                                           else dt[CV()]
14                                      end)
15     ) where dt is not null
16  /
Explained
Executed in 0,016 seconds
SQL> select * from table(dbms_xplan.display);
PLAN_TABLE_OUTPUT
Plan hash value: 1122788728
| Id  | Operation            | Name         | Rows  | Bytes |TempSpc| Cost (%CPU
|   0 | SELECT STATEMENT     |              |   100K|  2148K|       |   468   (4
|*  1 |  VIEW                |              |   100K|  2148K|       |   468   (4
|   2 |   SQL MODEL ORDERED  |              |   100K|   976K|       |   468   (4
|   3 |    WINDOW SORT       |              |   100K|   976K|  3944K|   468   (4
|   4 |     TABLE ACCESS FULL| AUDITLOG_TST |   100K|   976K|       |    53   (6
Predicate Information (identified by operation id):
   1 - filter("DT" IS NOT NULL)
16 rows selected
Executed in 0,125 seconds
SQL> No doubt that model gives best perfomance.