Author: Andrew Dados
Date: 09:37:50 02/21/00
Go up one level in this thread
On February 21, 2000 at 11:37:52, Mark Taylor wrote: >Can anyone help me with a tree searching problem? I am about to start my first >chess program soon, and have a question from working on my checkers program >over the last 10 years. > >The problem is that when the static evaluation was passed back up the tree, it >hardly changed the behaviour of the program, because the tree search showed that >that benefit could usually be gained "later" - so "good" moves did not get >played until the opponent was about to steal the advantage, then the program >would play the "right move" just in time! > >An idea I had was to have a small incrmental value subtracted from the eval, >this small increment getting larger the deeper into the tree search the eval was >returned from. I had already done this for the values WON & LOST, but I >hesitated doing it for positional evaluation because a) it seemed a debugging >nightmar; and b) it didn't seem correct "penalising" an eval the deeper search >it was returned from - intuitively deeper searches are to be trusted more than >shallow searches. > >What I did in the end I made the first search iteration look at positional eval >& material eval, then subsequent iterations looked at material eval only - but >this was really a cop out. > >What do other people do? > >Thanks, >Mark. I've seen this kind or behaviour in my program, too. However the deeper it searches, it usually 'discovers' that executing 'good moves' earlier gains more (or avoids trouble). So deeper search self-solves that oddity I think. BTW, an old chess proverb says that 'threat is stronger then its execution', so it looks like delaying taking that pawn can make some sense... -Andrew-
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.