Author: Uri Blass
Date: 04:38:58 03/31/02
Go up one level in this thread
On March 31, 2002 at 06:29:43, Gian-Carlo Pascutto wrote:
>On March 31, 2002 at 06:17:15, Klaus Friedel wrote:
>
>>Has anybody ever tried something like that in his null move code :
>>
>>
>> int nullDepth = depth - (NULL_REDUCE + 1)*DEPTH_BASE;
>>
>> if(tryNull){
>> executeNullMove();
>> beta -= NULL_BONUS;
>> value = - search(nullDepth, -beta, -beta+1, ply + 1);
>> undoNullMove();
>> if(value >= beta){
>> beta += NULL_BONUS;
>> ttStore();
>> return beta;
>> }
>> beta += NULL_BONUS;
>> }
>>
>>
>>Bigger values of NULL_BONUS increase the count of nodes prunded (but you migth
>>oversee some tactics). Values of about 30cp made my engine ply slightly better
>>than the default null-move (NULL_BONUS = 0).
>
>I tried this before, but no settings made my engine play stronger.
>
>The reason why this doesn't work very well is what I would call semi-zugzwangs.
>
>Say you have a program using piece-square tables, that does not consider
>attacked pieces in the evaluation.
>
>You have a knight on e5 that is attacked by a pawn. If you nullmove now,
>you are assuming that any move would be better than not moving. In reality,
>the best you can do is retreat the knight to f3 where it will get a smaller
>bonus from the piece-square tables. The NMO will fail, because not moving
>would have given you a higher score than moving.
>
>This is also the reason why side-to-move bonusses don't work very well in
>general.
>
>(With thanks to Christophe for explaining this once)
>
>--
>GCP
It is illogical to give a bonus for the side to move because
programs do not evaluate random positions.
A simple example is that they may evaluate this position when
black to move but they will never evaluate it when white to move
[D]4r1k1/2N2ppp/b7/8/8/5N1P/5PP1/6K1 b - - 0 1
It is possible that a bonus for the side that does not move
may be a better idea.
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.