Use analytic functions to group a set of records when timestamps in rows is less or equal than a value

Question

Hi all,

I have a table in an Oracle Database that has, among others, a DATE column that is loaded with the insertion timestamp of each row. I need to use existing data in such table to analyze the correlation between some events, so that with data like this:

COL_1         COL_2         TS
    A             1         Mon 15, February 2010 10:03:22
    B             2         Mon 15, February 2010 10:05:37
    C             3         Mon 15, February 2010 10:20:21
    D             4         Mon 15, February 2010 10:20:21
    E             5         Mon 15, February 2010 10:20:24
    F             6         Mon 15, February 2010 10:23:35
    G             7         Mon 15, February 2010 10:45:22

I would like to correlate to something like this, assumming related records are between a 5-minutes max difference between current and next "TS":

FIRST_TS                            COUNT
Mon 15, February 2010 10:03:22          2
Mon 15, February 2010 10:20:21          4
Mon 15, February 2010 10:45:22          1

Is is possible to use analytic functions to achieve this? How?

Answer 1

Hi Camilo JD,

This will group together rows that are less than 5 minutes distant from the previous row:

--ALTER SESSION SET nls_date_format= 'dy dd, month yyyy hh24:mi:ss';
--ALTER SESSION SET nls_date_language='ENGLISH';
SQL> WITH DATA AS (
  2  SELECT to_date('Mon 15, February 2010 10:03:22') ts FROM dual
  3  UNION ALL SELECT to_date('Mon 15, February 2010 10:05:37') FROM dual
  4  UNION ALL SELECT to_date('Mon 15, February 2010 10:20:21') FROM dual
  5  UNION ALL SELECT to_date('Mon 15, February 2010 10:20:21') FROM dual
  6  UNION ALL SELECT to_date('Mon 15, February 2010 10:20:24') FROM dual
  7  UNION ALL SELECT to_date('Mon 15, February 2010 10:23:35') FROM dual
  8  UNION ALL SELECT to_date('Mon 15, February 2010 10:45:22') FROM dual
  9  )
 10  SELECT MIN(ts) first_ts, COUNT(*) COUNT
 11    FROM (SELECT ts, SUM(gap) over(ORDER BY ts) ts_group
 12             FROM (SELECT ts,
 13                           CASE
 14                              WHEN ts - lag(ts) over(ORDER BY ts)
 15                                    <= 5 / (60 * 24) THEN
 16                               0
 17                              ELSE
 18                               1
 19                           END gap
 20                      FROM DATA))
 21  GROUP BY ts_group;

FIRST_TS                              COUNT
-------------------------------- ----------
mon 15, february  2010 10:03:22           2
mon 15, february  2010 10:20:21           4
mon 15, february  2010 10:45:22           1

Answer 2

I don't think you need analytics for this, you just need to generate some five minute intervals. The following code uses a Common Table Expression (AKA sub-query factoring) to generate five minute intervals from a given start date. The main query uses SUM() and CASE() to produce a count of records which fall within the interval

Here is the test data:

SQL> select * from t23
  2  /

C       COL2 COL3
- ---------- -----------------
A          1 15-feb-2010 10:03
B          2 15-feb-2010 10:05
C          3 15-feb-2010 10:20
D          4 15-feb-2010 10:20
E          5 15-feb-2010 10:20
F          6 15-feb-2010 10:23
G          7 15-feb-2010 10:45

7 rows selected.

SQL>

And here is the outcome

SQL> with t_range as ( 2 select to_date('15 February 2010 10:00','DD Month YYYY hh24:mi') 3 + ((level-1)/288) as this_5mins 4 , to_date('15 February 2010 10:00','DD Month YYYY hh24:mi') 5 + (level/288) as next_5mins 6 from dual 7 connect by level <= 12 8 ) 9 select t_range.this_5mins 10 , sum(case when t23.col3 >= t_range.this_5mins 11 and t23.col3 < t_range.next_5mins 12 then 1 13 else 0 end ) as cnt 14 from t23 cross join t_range 15 group by t_range.this_5mins 16 /

THIS_5MINS               CNT
----------------- ----------
15-feb-2010 10:10          0
15-feb-2010 10:20          4
15-feb-2010 10:30          0
15-feb-2010 10:05          1
15-feb-2010 10:55          0
15-feb-2010 10:15          0
15-feb-2010 10:40          0
15-feb-2010 10:45          1
15-feb-2010 10:00          1
15-feb-2010 10:35          0
15-feb-2010 10:25          0
15-feb-2010 10:50          0

12 rows selected.

SQL>

Answer 3

Here is a version with the analytic functions. Just substitute your table for the union subquery where I create a table with your data:

select distinct 
  first_value(ts) over (partition by continuous_group order by ts) first_ts
  , count(ts) over (partition by continuous_group) count
from (
  select col_1, col_2, ts, sum(discontinuity) over (order by ts) continuous_group
  from (
    select col_1, col_2, ts, case when lag(ts) over (order by ts) + numtodsinterval(5,'MINUTE') <= ts then 1 else 0 end discontinuity
    from (
    select 'A' col_1, 1 col_2, to_date('2010-2-15 10:03:22', 'YYYY-MM-DD HH24:MI:SS') ts from dual
    union (
    select 'B' col_1, 2 col_2, to_date('2010-2-15 10:05:37', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
    union (
    select 'C' col_1, 3 col_2, to_date('2010-2-15 10:20:21', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
    union (
    select 'D' col_1, 4 col_2, to_date('2010-2-15 10:20:21', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
    union (
    select 'E' col_1, 5 col_2, to_date('2010-2-15 10:20:24', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
    union (
    select 'F' col_1, 6 col_2, to_date('2010-2-15 10:23:35', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
    union (
    select 'G' col_1, 7 col_2, to_date('2010-2-15 10:45:22', 'YYYY-MM-DD HH24:MI:SS') ts from dual)
))
) order by first_value(ts) over (partition by continuous_group order by ts);

Answer 4

See here for a much more straightforward and reusable approach using analytic functions.