Author: Heiner Marxen
Date: 04:33:01 02/17/02
Go up one level in this thread
On February 16, 2002 at 17:51:22, Russell Reagan wrote: >What determines the "correct" or "best" move in any given position? > >In addition to your own definition of what defines the "correct" or "best" move, >I'd like to know people's thoughts on the following questions: > >Is the correct move the move that leads to the highest evaluation, namely win, >loss, or draw? In theory: yes. Of course, most of the time we do not know the theoretic value. Also note, that the "best" need not be a single move, it is a non-empty set of moves. The "bm" opcode of EPD therefore allows a list of moves. >Does the number of moves taken to achieve the highest evaluation matter? For >example, is a move that leads to a mate in 5 any better than a move that leads >to a mate in 6 (assuming the 50 move rule would not affect the outcome of the >game given the additional move)? That depends :-) If this is the first mate you recognize (in the current playing line) then any mate is "best" insofar as it guarantess a win (however deep). If one move before you already followed a move because it was recognized as a mate-in-6, now another mate-in-6 is not enough, since you do not make progress. You should insist to find a mate-in-5 (it must be there). >If you have two moves that lead to a mate in X, how do you determine which of >them are the better move? Does anyone take any additional considerations into >their program to account for a position of the following nature: Let's say that >your program is playing a grandmaster and your program realizes that the GM has >a mate in 10 in a complicated position. In reality, there are two moves worth >considering for your program. Move A leads to a mate in 9 for the GM, and move B >leads to a mate in 10 for the GM. Move A, which leads to the mate in 9 is much >more complex and there are many attractive moves at each step of the way for the >GM to play that let your program escape mate and acheive a draw farther down the >road. Move B, which leads to a mate in 10 for the GM, is pretty straight forward >and the GM should be able to tighten the noose and finish off your program in 10 >more moves. I think (possibly incorrectly) that it is best to go with move A >which makes it very hard for the GM to win, although not impossible. I have a >feeling that most computer programs are going to pick the mate in 10 route >though. Any thoughts or actual methods used to handle this kind of thing in your >program? Here you argue how to handle partial knowledge, a somehow imperfect opponent in the context of a special program. Hence, we have a lot of freedom to model the knowledge of the opponent (which will be guesswork), and to predict how well the current program will handle this or that position. Within the alpha-beta (or mini-max) framework, the opponent is modelled by the program itself (we expect him to choose whatever we would choose ourselves). The "contempt factor" used by some programs is a modification of this, but of course much more complex models of an opponent are possible. But I've not yet heard/read any more complex models of opponents. I've no personal experience with this topic, so I will stop talking here. >Russell Cheers, Heiner
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.