Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Scatha 0.2.6 released

Author: Alessandro Scotti

Date: 23:49:30 10/03/05

Go up one level in this thread


On October 04, 2005 at 02:35:29, Ross Boyd wrote:

>I've had a user request for this feature. To be honest, it goes against the
>grain after toiling so hard for a handful of elo here and there. However, I plan
>to honour the request when T2 is released which is still a long way off yet.
>
>As you indicate, its not as simple as randomly pruning moves from ply 2.
>
>My user suggested two approaches.
>1. Make the search depth random between two adjustable depths- for example 3ply
>and 6ply. Apparently this method is used in Crafty SE...
>2. Make the engine choose ply 1 moves randomly but within a certain score range
>of the best move... so it would occasionally play poor moves that lead to lower
>positional scores without throwing pieces away.

Hi Ross,
that's a possible approach, but I don't like to throw away information. Instead,
it should be possible to search all root moves at full depth and then pick one
according to its score *and* PV.
So if engine sees its score go up thanks to a tactical shot that gains material,
it can "overlook" that continuation and pick another line. Conversely, if it
finds a line where it goes down a piece with a capture that happens a couple of
moves down the PV it can play that and see if the opponent takes the chance.
There are other possibilities as well, to use for evaluation, e.g. leaving king,
rooks and queen in "forkable" positions, downplaying king safety a bit and so on
in order to increase the chance that a combination can be found... this should
help in getting fun I think.



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.