Author: scott farrell
Date: 20:54:29 01/09/03
Go up one level in this thread
On January 09, 2003 at 14:17:06, Stan Arts wrote: >Hello, > >I often read computerprograms are so bad at correspondence chess. That would >make sence, since they run up a very steep wall after about 10-15 full moves of >analysis. > >Wouldn´t it be possible however to write a program, that could spend it´s hours/ >days of time for a move in a different way, searching much shorter, say just 5 >minutes or less per move and then automaticly "play" (like in normal games) >lines against itself to great depths, sometimes discovering refutations, and >then disregarding this move and try to resolve other moves this way. > >So it would be sort of like replacing the human doing the analysis with computer >help by another part of computer-code. > >Wouldn´t it be possible for the program to get to greater "depths" than it can >in >normal search? And find much deeper ideas and refutations like this? > >BUT, there are no programs doing this..so that would usually mean this idea >doesn´t work. :) > >Why wouldn´t this work? > >Well, just a thought i have been having, > >Stan I think a different search could be employed. Mate solvers use different search techniques to standard engines. I think "singular extensions" might be useful in corro games. This is where one move picked as being pretty good, and searched much more deeply than the rest. This is in essense similar to your, "play the game" idea, but a little smarter. You might also pick a good move after a few hours, and then go really deep trying to prove there is no trap, rather than also trying to find a better move, which is way way quicker for computer to do. I think really tight aspiration windows could help also. Scott
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.