Need a maximum date value using group by

Question

Create table student (dept number(10), dep_name varchar2(10),join_date date,years_attended number(10),end_date date);
insert into student values (1,'I',to_date('3/7/1917','MM/DD/YYYY'),4,to_date('8/26/1987','MM/DD/YYYY'));
insert into student values (1,'I',to_date('1/1/1900','MM/DD/YYYY'),4,to_date('8/26/1932','MM/DD/YYYY'));
insert into student values (1,'D',to_date('1/1/1920','MM/DD/YYYY'),5,to_date('8/26/1994','MM/DD/YYYY'));
insert into student values (1,'C',to_date('1/1/1920','MM/DD/YYYY'),6,to_date('8/26/1945','MM/DD/YYYY'));
insert into student values (2,'I',to_date('7/1/1900','MM/DD/YYYY'),3,to_date('8/26/1932','MM/DD/YYYY'));
insert into student values (2,'I',to_date('8/16/1916','MM/DD/YYYY'),9,to_date('8/26/1923','MM/DD/YYYY'));
insert into student values (2,'D',to_date('8/16/1916','MM/DD/YYYY'),10,to_date('8/26/1987','MM/DD/YYYY'));
insert into student values (3,'I',to_date('3/7/1917','MM/DD/YYYY'),4,to_date('8/26/1987','MM/DD/YYYY'));
insert into student values (3,'D',to_date('7/28/1920','MM/DD/YYYY'),6,to_date('8/26/1945','MM/DD/YYYY'));
insert into student values (3,'I',to_date('7/28/1920','MM/DD/YYYY'),8,to_date('8/26/1965','MM/DD/YYYY'));
insert into student values (4,'I',to_date('12/31/1924','MM/DD/YYYY'),2,to_date('8/26/1998','MM/DD/YYYY'));
insert into student values (4,'I',to_date('6/10/1929','MM/DD/YYYY'),1,to_date('8/26/1943','MM/DD/YYYY'));
insert into student values (4,'C',to_date('1/17/1927','MM/DD/YYYY'),4,to_date('8/26/1955','MM/DD/YYYY'));
insert into student values (4,'C',to_date('6/10/1929','MM/DD/YYYY'),30,to_date('8/26/1967','MM/DD/YYYY'));
insert into student values (5,'D',to_date('2/10/1931','MM/DD/YYYY'),2,to_date('8/26/1943','MM/DD/YYYY'));
insert into student values (5,'I',to_date('2/10/1931','MM/DD/YYYY'),24,to_date('8/26/1962','MM/DD/YYYY'));
commit;I need a maximum date value join_date for each department. If max(join_date) has two records for each dept then max(end_date) should be considered. I have used a below select query
select * from student where join_date in (select
max(join_date) from student group by dept);which gives me the following result
1     D     1/1/1920     5     8/26/1994
1     C     1/1/1920     6     8/26/1945
2     I     8/16/1916     9     8/26/1923
2     D     8/16/1916     10     8/26/1987
3     D     7/28/1920     6     8/26/1945
3     I     7/28/1920     8     8/26/1965
4     I     6/10/1929     1     8/26/1943
4     C     6/10/1929     30     8/26/1967
5     D     2/10/1931     2     8/26/1943
5     I     2/10/1931     24     8/26/1962But I am looking for the result which gives me only one maximum value for each dept column. First it should look for maximum value of join_date, if two records has same join_date then max(end_date) should be considered. The result should be sumthing like this
1     D     1/1/1920     5     8/26/1994
2     D     8/16/1916     10     8/26/1987
3     I     7/28/1920     8     8/26/1965
4     C     6/10/1929     30     8/26/1967
5     I     2/10/1931     24     8/26/1962Can you please tell me how to rewrite the select query to get the above results.
Edited by: user11872870 on Aug 2, 2011 5:29 PM
Edited by: user11872870 on Aug 2, 2011 5:36 PM

Answer 1

Hi,
That's called a Top-N Query , and here's one way to do it:
WITH     got_r_num     AS
     SELECT     student.*
     ,     ROW_NUMBER () OVER ( PARTITION BY  dept
                               ORDER BY          join_date     DESC
                         ,                end_date     DESC
                       )      AS r_num
     FROM    student
SELECT       dept, dep_name, join_date, years_attended, end_date
FROM       got_r_num
WHERE       r_num     = 1
ORDER BY  dept
;Another way is similar to what you originally posted:
SELECT    *
FROM       student
WHERE        (dept, join_date, end_date)
               IN (
                    SELECT    dept
               ,       MAX (join_date)
               ,       MAX (end_date) KEEP (DENSE_RANK LAST ORDER BY join_date)
               FROM      student
               GROUP BY     dept
               );I suspect the first way (using ROW_NUMBER) will be faster.
Also, the ROW_NUMBER approach is guaranteed to return only 1 row per dept. Using the GROUP BY approach,if there is a tie on join_date and end_date, then it will return all contenders in that dept. Using ROW_NUMBER, it's easy to add as many tie-breaking expressions as you want, and, if there is still a tie, it will arbirarily pick one of the rows involved in the tie as #1.
Thanks for posting the CREATE TABLE and INSERT statments! That's very helpful.
Edited by: Frank Kulash on Aug 2, 2011 9:00 PM
Added GROUP BY alternative