Author: Dieter Buerssner
Date: 18:04:42 07/11/04
Go up one level in this thread
On July 11, 2004 at 20:18:02, Stuart Cracraft wrote:
>I guess one might want a similar randomness for the program to pick amongst
>equally similar moves +- a few evaluation points.
I'd fear, that adding some +- to the return of eval could hurt the efficiency of
transposition tables rather badly. But I am not sure. An approach, that should
not hurt could be, to randomize some eval constants *once* each game. Say piece
square tables. Or something like bonus_for_rook_on_7th_rank. If you use 20 or 22
for such a bonus should not change the strength of the engine. But changing
several such values should at least make it rather improbable that the engine
will be very deterministic from one game to the other ("I can win all the time
with exactly this variation"). Still, one might need to be careful, to keep the
eval constants consistent. Say you have passed_pawn_on_4th_rank=15 and
passed_pawn_on_5th_rank=17 (perhaps not an realistic example, just to show my
point). Now, you want to add a random factor of +-2 to each value.
passed_pawn_on_4th_rank might end up as 17 and passed_pawn_on_5th_rank as 15.
And the engine might refuse to advance its passed pawns.
>No nuclear simulations here.
An off topic answer in another post.
Regards,
Dieter
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.