Author: Gunnar Andersson
Date: 05:58:22 10/29/98
Go up one level in this thread
On October 27, 1998 at 15:30:12, Dann Corbit wrote: >On October 27, 1998 at 09:09:46, Gunnar Andersson wrote: >[snip] >This is very interesting. I think that there is still a problem with applying >this to chess, but I would be interested to hear your thoughts. The biggest >problem with chess search/evaluation functions (as I see it) is that the >information is almost *never* perfect. There are so many combinations and >possibilites that you are always relying on an estimate. And you can never >carry out the calculations to the end to find out either. >How would this technique work when the information is known to be imperfect >(sometimes completely wrong!)? The imperfectness of the information is due to the (lack of) quality of the evaluation function. The leaf nodes in the minimax tree could in chess correspond to checkmates or resigned positions. Positions where the players agree on a draw are more problematic, I don't know how to handle these cases - this does not arise in Othello. I've had the following experiences with this approach for Othello: Sometimes the evaluation function+search misjudges some sharp line, and for a while the program will make deviations which look good but prove not-so-good after a few games. Usually this will make the program avoid these lines after having tried them until they can't find any good deviations. The deviations are calculated with searches to about 16-18 plies (8-10 of which are full-width). / Gunnar
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.