Selecting top n elements of a group in Oracle

Question

I have an Oracle table which has a name,value,time columns.Basically the table is for logging purposes to store what are the changes made to a particular name,what was the previous value and what time the change was made.

I need to formulate a query to fetch the top n changes for a particular name,and the output should have all the names in the table. Any help/suggesstions?

Edit:

Name         Value     Time
Harish       Pass      1-Nov-2011
Ravi         Fail      2-Nov-2011
Harish       Absent    31-Oct-2011
Harish       Attended  31-Aug-2011 
Harish       Present   31-Jul-2011

I need to select details of Harish on 1st Nov,Oct 31st,31st Aug and Ravi.

Answer 1

Is this what you are after?

My test set-up:

SQL> alter session set nls_date_format = 'DD-Mon-YYYY HH24:Mi:SS';

Session altered.

SQL> drop table so_test;

Table dropped.

SQL> create table so_test  (
  2    n varchar2(32)
  3  , v varchar2(32)
  4  , t date );

Table created.

SQL> 
SQL> insert into so_test values ( 'X' , 'Test1', to_date('01-Jan-2011 12:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> insert into so_test values ( 'X' , 'Test2', to_date('01-Jan-2011 13:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> insert into so_test values ( 'X' , 'Test3', to_date('01-Jan-2011 14:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> insert into so_test values ( 'Y' , 'Test5', to_date('02-Jan-2011 12:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> insert into so_test values ( 'Y' , 'Test6', to_date('03-Jan-2011 12:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> insert into so_test values ( 'Y' , 'Test7', to_date('04-Jan-2011 12:00:00','DD-Mon-YYYY HH24:Mi:SS') );

1 row created.

SQL> 

Here is the query:

SQL> select n,v,t from (
  2  select n, v , t , rank() over ( partition by n order by t desc) r
  3  from so_test
  4  ) where r <= 2;

N                V                T
-------------------------------- -------------------------------- --------------------
X                Test3                01-Jan-2011 14:00:00
X                Test2                01-Jan-2011 13:00:00
Y                Test7                04-Jan-2011 12:00:00
Y                Test6                03-Jan-2011 12:00:00

SQL> 

Answer 2

select * from (select name, value, 
time, ROW_NUMBER OVER (PARTITION BY name ORDER BY name) change_no
from table )
where change_no <= 100 AND name ="abc"
ORDER BY TIME

Assuming name remain same, and changes are made to the "value".

Answer 3

Matthew Watson's answer is not always valid, if ordering column is duplicated, query returns more than "r" rows. The solution is to concatenate an unique value to ordering column, it can be used a primary key of the table. Example:

SELECT * FROM (
  SELECT 
      t.*,
      RANK() OVER (PARTITION BY object_type ORDER BY (to_char(created,'YYYYMMDDHH24MISS') || object_id) DESC) rank
    FROM ALL_OBJECTS t
    ) 
    WHERE rank <= 3

Answer 4

I offer my quick fix. This query groups and counts in descending order (in the inner query). The Outer query simply allows you to define the number of rows to display (:pRows).

Select * from
(Select
     group_field,
     count(*) Cnt
From
     record_source(s)
Where
     Conditions
Order by
     Count(*) Desc) x
Where
  rownum < :pRows+1;

Answer 5

select name, VALUE, TIMESTAMP 
from (select name, VALUE, TIMESTAMP, rank() over (partition by NAME order by TIMESTAMP DESC) rank 
     from logs) 
where rank <= 3