Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SEEing it is BLIND

Author: Tord Romstad

Date: 06:24:46 08/09/04

Go up one level in this thread


On August 08, 2004 at 19:50:15, José Carlos wrote:

>  Another idea that seems to work for me (in my non-released private engine,
>where I evaluate internal nodes) is to consider the king safety changes along
>the path. A single move is not likely to make a big change (unless some crazy
>sacrifices in king's pawn shelter), but some manoeuvres do the trick. I don't
>recall the details, but the idea goes like this:
>
>if (eval[ply] > eval[ply-1] + MARGIN1)
>  extend;
>if (eval[ply] > eval[ply-2] + MARGIN2)
>  extend;
>if (eval[ply] > eval[ply-3] + MARGIN3)
>  extend;
>
>  Where MARGIN3 > MARGIN2 > MARGIN 1, and eval means king safety evaluation of
>the opponent. Or maybe I used full moves (ply-1,ply-3,ply-5), I'm not sure. It's
>been a long time without doing chess programming.
>  But the idea worked very well in tests, and made the engine play interesting
>games also.

I've thought about similar ideas, but never tried them out in practice.
If they work for you, I should probably give it a try as well.  Thanks
for the encouragment!

Another idea, which I have tried from time to time but never decided
whether I like or not, is what I call "compensation extensions".  These
are triggered when one side has a big material advantage, but the other
side has a sufficiently big positional advantage that the static eval
returns a score close to 0.  This usually happens in very wild positions,
where the static eval tends to be extremely inaccurate.  The position
will most often be lost for one of the players, despite the static eval
near 0.  It makes some sense to extend in such positions.

Currently, I use this idea only in a limited way.  My attack extensions
and mate threat extensions are bigger when the attacking side is behind
in material than when the material is equal or the attacking side is ahead.

Tord



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.