SQL to find grouped rows which have changed values?

Question

Hi,
I am wondering if there is an efficient pure SQL way to do this...?
I have a table that has a number of columns:
InsertionDate
Indentity
Col-a to Col-z - some data.
the pair (InsertionDate, Identity) are unique.
The rows represent the same values as they change over time, for example:
Select * from <mytable> order by Identity, InsertionDate;
InsertionDate Identity Balance
10-Feb-01 1001 -100
11-Feb-01 1001 1
12-Feb-01 1001 50
10-Feb-01 1002 10
11-Feb-01 1002 10
12-Feb-01 1002 76
I want to do the following:
Select from <mytable> where InsertionDate and the next InsertionDate have a different value for Balance 'grouped by' Identity and InsertionDate = "10-Feb-01";
[I use 'grouped by' very cautiously - I cannot explain in "pseudo sql"  without using the misleading grouped by term]
To produce the output:
InsertionDate Identity Balance
10-Feb-01 1001 -100
11-Feb-01 1001 1
Select from <mytable> where InsertionDate and the next InsertionDate have a different value for Balance 'grouped by' Identity and InsertionDate = "11-Feb-01";
InsertionDate Identity Balance
11-Feb-01 1001 1
12-Feb-01 1001 50
11-Feb-01 1002 10
12-Feb-01 1002 76
There may be many columns in addition to Balance which also need to be checked for changes.
Cheers,
Karl.

Answer 1

Hi, Karl,
One way is to self-join two copies of the table, so you can compare any number of columns from adjacent rows.
For example:
WITH     got_r_num     AS
     SELECT     insertiondate, identity, balance
     ,     ROW_NUMBER () OVER ( PARTITION BY  identity
                               ORDER BY          insertiondate
                       ) AS r_num
     FROM     mytable
SELECT     a.insertiondate, a.identity, a.balance
FROM     got_r_num      a
JOIN     got_r_num      b     ON     a.identity     = b.identity
                     AND     a.r_num          IN ( b.r_num + 1
                                          , b.r_num - 1
WHERE     a.balance     != b.balance
AND     TO_DATE ( '11-Feb-2001'     -- Use a bind variable if you don't want to hard-code this parameter
          , 'DD-Mon-YYYY'
          ) = LEAST ( a.insertiondate
                     , b.insertiondate
ORDER BY  identity
,            insertiondate
;Here, table a is the one to be displayed, and table b contains either the previous or the next row with the same identity.
By the way, whenever you post a question, it really helps if you provide the sample data in a form that people can use to re-create the problem and test their solutions. CREATE TABLE and INSERT statements are best, for example:
CREATE TABLE     mytable
(      insertiondate     DATE
,      identity          NUMBER (4)
,      balance                 NUMBER
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('10-Feb-2001', 'DD-Mon-YYYY'), 1001, -100);
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('11-Feb-2001', 'DD-Mon-YYYY'), 1001, 1);
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('12-Feb-2001', 'DD-Mon-YYYY'), 1001, 50);
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('10-Feb-2001', 'DD-Mon-YYYY'), 1002, 10);
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('11-Feb-2001', 'DD-Mon-YYYY'), 1002, 10);
INSERT INTO mytable (insertiondate, identity, balance) VALUES (TO_DATE ('12-Feb-2001', 'DD-Mon-YYYY'), 1002, 76);
COMMIT;I'm not sure what you mean by "There may be many columns in addition to Balance which also need to be checked for changes."
Why not post an example that has 1 or 2 columns like balance, and the results you want for that.
At first, this problem sounds like a job for the analytic LAG and/or LEAD functions, but depending on your answer to the last question, that may be very tedious: you'd need a separate function call (maybe even two) for each column, because a function can only return one value, but a join gets all the values at once.
Edited by: Frank Kulash on Jun 11, 2010 10:43 AM