Oracle Group By

Question

Hello all!
I've a question that I saw a lot of discution but it doesn't came to the solution I need, so if someone could give me a light I'd appreciate that!
The problem is: I need to do a query that returns some values, but one of the column can't be on the GROUP BY clause. Let's try to explain the cenario:
Suppose I have a table of registry with these infos:
GPS   Date                          Event_id
*01 21/10/2012 09:08:12 100*
*02 22/10/2012 10:00:05 105*
*05 19/10/2012 00:00:00 120*
*03 22/10/2012 17:53:10 100*
*01 22/10/2012 07:05:14 125*
*01 21/10/2012 21:38:00 110*
*02 18/10/2012 03:15:17 115*
I'd like to have the "most recent" registry of each GPS but also the event_id that occured ont that registry.
If I use the query
SELECT gps, MAX(Date)
FROM table
GROUP BY gps
ORDER BY gps
Then I get the result I want, but without the event_id
GPS    Date
*01 22/10/2012 07:05:14*
*02 22/10/2012 10:00:05*
*03 22/10/2012 17:53:10*
*05 19/10/2012 00:00:00*
I'd like to have respectively the info
Event_id
*125*
*105*
*100*
*120*
But, if I put the event_id column on the SELECT, I have to put it on the GROUP BY to it works. If I do so, the result I get is pretty the same as the whole table because of the different event_ids I have for each registry.
Does someone have any idea on how do I do this?
By the way, I'm using Oracle 11g
Thanks in advance,
Bruno.

Answer 1

Hi, Bruno,
Welcome to the forum!
Here's one way:
SELECT    gps
,        MAX (dt)          -- DATE is not a good column anme
                     AS max_dt
,       MIN (event_id) KEEP (DENSE_RANK LAST ORDER BY dt)
                     AS last_event_id
FROM      table_x          -- TABLE is not a good table name
GROUP BY  gps
ORDER BY  gps
;What if 2 (or more) rows with the same gps have exactly the same date? That may not be possible in your table, but if it does happen, then the query above wouold chose the lowest of the event_ids from the latest date; that's what MIN (event_id) is doing.
For example, if this were your sample data:
GPS     DT                EVENT_ID
05      01/01/2012 00:00:00   102
05      19/10/2012 00:00:00   120
05      19/10/2012 00:00:00   121There is a for the last row. Both event_id=120 and event_id=121 have an equal claim to being the last row. In this case, the query above would return 120 as the last_event_id, because it is the lowest number from among the contenders. (Note that this is not the same as the lowest event_id overall. 102 is the lowest event_id, but it has no claim to being the last row.)
If you want to pick all the contenders in case of a tie (that is, both 120 and 121, given the data above), then you could do soemthing like this:
WITH     got_r_num     AS
     SELECT  gps, dt, event_id
     ,     RANK () OVER ( PARTITION BY  gps
                           ORDER BY          dt     DESC
                    )           AS r_num
     FROM    table_x
SELECT    gps, dt, event_id
FROM       got_r_num
WHERE       r_num     = 1
ORDER BY  gps, event_id
;Whenever you have a question, please post CREATE TABLE and INSERT statement for your sample data. Also post the exact resuolts you want from that data, exactly as you want to see them.
See the forum FAQ {message:id=9360002}
Edited by: Frank Kulash on Oct 22, 2012 3:40 PM
Edited by: Frank Kulash on Oct 22, 2012 3:46 PM