I have an ACCOUNT table with columns ACCT_NBR, STYLE and PERCENTAGE. There can be many records for a single account (that is, there can be many records with same ACCT_NBR column value). I need to select all accounts having records similar to a given account.
AccNum1 Style1 Percentage1 AccNum1 Style2 Percentage2 AccNum2 Style1 Percentage1 AccNum3 Style1 Percentage1 AccNum3 Style2 Percentage2
Here, AccNum1 and AccNum3 have the same values in columns (other than ACCT_NBR column). So, I need to collect the list of similar accounts together for a big batch process. Could you please let me know how to achieve it using analytical functions? I have tried and done it using SQL.
Right off the top I must apologize, because this solution will not use analytical functions. Sorry, I'm just not up to speed on them yet. I don't even know if a solution to your question is possible using analytical functions. Hopefully, one of my five or six regular readers will know, and send in a query.
The way to solve this with ordinary SQL is with a grouped self-join.
select t2.ACCT_NBR from ACCOUNT as t1 inner join ACCOUNT as t2 on t1.ACCT_NBR <> t2.ACCT_NBR and t1.STYLE = t2.STYLE and t1.PERCENTAGE = t2.PERCENTAGE where t1.ACCT_NBR = 'AccNum1' group by t2.ACCT_NBR having count(*) = ( select count(*) from ACCOUNT where ACCT_NBR = 'AccNum1' )
This self-join query joins each style/preference row for AccNum1 (see WHERE clause) with every matching style/preference row for any other account (except itself), and groups the result by the other account. If the number of matching rows for the other account is the same as the number of rows for AccNum1, then the other account satisfies the HAVING clause, and is returned in the result set.
In the HAVING clause, you can see a subquery. This is an uncorrelated subquery, since it is not related to each particular GROUP BY group, which in this case is T2.ACCT_NBR. This means the subquery can be evaluated once, before the main query starts execution. Then the main query produces a COUNT(*) for each T2.ACCT_NBR, which is then compared to the subquery count obtained earlier. If there are any missing style/preference rows for the other account, then its COUNT(*) will be less than the count for AccNum1. Neat, eh?
This was first published in October 2005