Author: David Blackman
Date: 02:45:19 05/03/99
Go up one level in this thread
On May 03, 1999 at 05:04:47, Francis Monkman wrote: >The idea is more or less along these lines: keep the score for each move at each >ply-depth of search, then the resulting 'statistical curve(s)' can be used to >determine whether a move is gaining in 'expectation-value' (and hence should >definitely be explored thoroughly). Conversly, programs often seem to choose >moves that seem stronger,even though they are losing expectation-value as the >search deepens -- perhaps this might be avoided by such a method, too. > >Francis This is a bit tricky with alpha-beta because you don't get exact evaluation for anything except the "best" move. Hence the idea of using node counts. Any other ideas for finding 'expectation-value'?
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.