Ya,

It isn't enough, that is why I labeled it a difficult problem. Consider

x y A001 1 A001 3 A002 1 A002 2 A003 1 A003 2

You have to choose

A001 3

in order to win this one, but you cannot know that until after you have made the choices for A002 and A003. If you do it in a DATA step, I think you need all the records in an array and enough book keeping to back track when you have made a bad choice, and then to use that bad choice if it happens to be a bad choice.

In other words, to do it in a DATA step, you need a programmer like Paul Dorfman.

On the other hand, there was a panel of programmers at the last SGF that wrote programs to solve SUDOKU puzzles and this seems to require the same sort of logic.

Ian Whitlock ============== Date: Tue, 18 Dec 2007 14:23:58 -0500 Reply-To: Ya Huang <ya.huang@AMYLIN.COM> Sender: "SAS(r) Discussion" From: Ya Huang <ya.huang@AMYLIN.COM> Subject: Re: selecting a unique set of data. Comments: To: Ian Whitlock <iw1junk@COMCAST.NET>

Thanks Ian,

I thought about sorting the data by the number of unique y in each x, so that, the least number of y group get to pick the y first.

This could be done easily by SQL:

proc sql; select *,count(distinct y) as ny from have group by x order by ny,x,y ;

But then I wasn't sure this is enough, so I didn't post.

<SNIP>