Author: Christophe Theron
Date: 17:52:52 12/03/02
Go up one level in this thread
On December 02, 2002 at 15:40:14, Jim Bumgardner wrote:
>Has anyone experimented with the following ideas or similar ones? What
>was your experience? Is it sensible to modify your strategy when
>playing against humans versus computers?
>
>* * *
>
>Given two moves which are roughly equal (within 1/8th pawn or so):
Oops... You'd better stop here.
Alphabeta and all the efficient known tree search algorithms (SSS*, MTD(f) and
others) are all able to find a best move, but absolutely unable in general to
tell you what the second-best moves are, and what the score differences with the
best move are.
So you don't have that basic information about "moves which are roughly equal"
unless you are ready to slow down these algorithms by a factor of 100 or more,
which would make your program much weaker.
I think the only sensible way to deal with faillible opponents is to tweak the
evaluation function, and maybe to prune more moves when it is the faillible
opponent's turn (but I have not tried yet).
Everything that deals with "roughly equal moves" or "number of refutations" and
the like is not realistic. The constraint here is to use only the information
that alphabeta can produce.
Christophe
> a) Choose the move which had more opponent replies which looked
> good but dropped off deeper in the search, hoping the opponent
> will choose a bad reply.
>
> b) Choose the move which has the fewest 'good' replies (forcing lines),
> hoping the opponent will miss them.
>
> c) Choose the move for which there are fewer forcing lines for us (e.g.
> for which we have more replies to the opponent's higher-scoring moves).
> This is the compliment of "b".
>
> d) Choose the move for which the opponent's line isn't forced (the opposite
> of "b") hoping to get him into a complex situation which he can't handle.
>
> e) Choose the move which wasn't part of the PV the last ply, especially if
> the move looked bad the previous ply (hoping the opponent has been only
> thinking about the PV - might be good for playing against pondering
> computers to get them to squander their pondering).
>
>- jbum
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.