Re: computer-go: Re: Need help building Graded Gnu Go Problem Database

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

I know  from my work  on computer  chess  that problem sets  are quite
useful, but  you can't take them  too seriously either.  For instance,
no  one has figured  out how to  build a set that  will tell you which
program is better.

Still, a good  problem  set is  especially useful for  debugging.  You
will find many bugs  by observing how  a problem affects the solutions
of various problems.

If you  use  a problem  set  to improve  your   program,  you can  get
spectacular  results  in  the number of   problems  you solve  without
actually improving your program.  Most  peoples intuitition is that if
you  make a  big  improvement on some   problem  set, you  will almost
certainly improve  your program too.   This may or  may not be true so
don't get too excited.

Don




   X-Sender: fotland@xxxxxxxxxxxxxxxxx
   X-Mailer: QUALCOMM Windows Eudora Version 5.1
   Date: Mon, 30 Jul 2001 09:12:42 -0700
   From: David Fotland <fotland@xxxxxxxxxxxxxxxxx>
   References: <200107290237.TAA31377@xxxxxxxxxxxxxxxxx>
   Mime-Version: 1.0
   Sender: owner-computer-go@xxxxxxxxxxxxxxxxx
   Precedence: bulk
   Reply-To: computer-go@xxxxxxxxxxxxxxxxx
   Content-Type: text/plain; charset="us-ascii"; format=flowed
   Content-Length: 1600


   Wolf's problems are great for general reading skills, but they are usually 
   very unlike positions that
   come up in actual games, so I recommend that his problems be a small part 
   of any computer go
   test set.

   David

   At 11:57 AM 7/30/01 +0200, you wrote:
   >Pierce wrote:
   >
   > > So I asked Thomas Wolf the author of GoTools if he would be willing to
   > > make a small fraction of his 24,000 problem set available as part of a
   > > public domain benchmark for go program developers.
   > >
   > > He agreed! He's willing to make somewhere between 500-1000 problems
   > > available. Which is really cool, as the computer-go world will be able
   > > to have a standard set of problems they can exchange. I'll put them
   > > under the LGPL (Library Gnu Public License or "Lesser" GPL) so its clear
   > > that by including them your program doesn't be come public too...
   >
   >First of all, thanks to Pierce and Thomas Wolf for making this possible!
   >
   >One suggestion: the problems should be split upfront into two (statistically
   >matched) sets, one to be used in Go program development, the other reserved
   >for validation of the "final" program.  While no panacea, this should help
   >somewhat to prevent the development of programs specialized to solve only
   >this particular set of problems...
   >
   >Best wishes,
   >
   >--
   >     Dr. Nicol N. Schraudolph              http://www.icos.ethz.ch/~schraudo/
   >     Institute of Computational Sciences            mobile:  +41-76-585-3877
   >     ETH Zentrum, WET-D, Weinbergstr. 43            office:     -1-632-7942
   >     CH-8092 Zuerich, Switzerland                      fax:           -1703