Question

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

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