Author: martin fierz
Date: 15:57:23 03/19/04
Go up one level in this thread
On March 19, 2004 at 18:16:45, Uri Blass wrote:
>On March 19, 2004 at 17:39:17, Dieter Buerssner wrote:
>
>>I think, the learning issue was solved. I suggest to go further. For Crafty to
>>be a good sparring partner (this might depend on the version of Crafty), you
>>should make it not aware, that the opponent is a computer (can be done, by
>>options to WB/XB). Otherwise, Crafty will always use the same book lines ("My"
>>Crafty that I downloaded will always make the same move in each position, when
>>the opponent is a computer. It will chose the move with the best statistics in
>>the PGN from which the book was created). If your book is rather random, this
>>might be acceptable, but if you also try with a rather forced book, you will see
>>the same variations - possibly the same games - over and over again. Even when
>>your book is rather broad, this might not be, what you expect from a sparring
>>partner. Of course, you should not mention your results against Crafty, when you
>>disabled learning and the narrow book - Bob will complain loudly :-)
>>
>>Regards,
>>Dieter
>
>I think that martin does not care about deterministic behaviour because he is
>using nunn match conditions for testing.
>
>The problem is that he likes to compare a version with future versions so if
>Crafty behaves different because of learning it is a problem.
exactly! learning is nice of course, but if you want to measure progress it's a
pain if your opponent learns, specially if you keep running matches from the
same starting positions...
cheers
martin
>Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.