Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: High branching factor games

Author: Bob Durrett

Date: 06:04:42 11/27/02

Go up one level in this thread


On November 27, 2002 at 05:23:56, Uri Blass wrote:

>On November 27, 2002 at 04:58:36, Russell Reagan wrote:
>
>>On November 27, 2002 at 03:52:37, Ricardo Gibert wrote:
>>
>>>Consider backgammon. Despite the element of chance, alpha beta is still
>>>applicable. Also, because if the element of chance, the game has a very high
>>>branching factor. You must account for all possible rolls in addition to all the
>>>possible legal moves with the checkers.
>>>
>>>The best backgammon programs play better than the best humans despite the high
>>>branching factor. Humans do not really try to calculate all the possibilities,
>>>so a program that searches just 3 ply ahead can outplay strong humans.
>>>
>>>The other oddity about backgammon is that the use of neural nets has actually
>>>been effective for evaluation. This contrasts with the results in chess.
>>>
>>>If you are interested in reading more about this you can check out the following
>>>site: http://www.bgsnowie.com/snowie/snowie.dhtml
>>
>>
>>So is it fair to say that backgammon, despite having a high branching factor, is
>>still playable by computers because it is a relatively simple game? For example,
>>in backgammon, you say 3 ply is sufficient. Go programs can get 3 ply, but it's
>>far from sufficient.
>
>The number of plies is not the important factor.
>
>In chess computers can get more plies in the endgame but it does not help them
>to be relatively better against humans in the endgame.
>
>I think that the main problem in go is that it is hard to define the evaluation
>function.

Maybe that's the key.  Dispense with the search algorithm entirely, and come up
with a killer evaluation function.

In theory, that would work in chess too.  All you need is a perfect evaluation
function to evaluate the initial position of the game.  : )

Bob D.


>
>I expect chess programs to get at least master level in chess even if you
>increase the size of the board to 100*100.
>
>Humans may have big problems in this game because it may take them time to
>calculate if a move is legal.
>
>in 8*8 the eye can see immediately that the rook cannot go from c1 to d8
>in 100*100 board it is not going to be trivial to see that the rook cannot go
>from y27 to z78
>
>Uri



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.