Author: Robert Hyatt
Date: 18:17:10 09/26/01
Go up one level in this thread
On September 26, 2001 at 17:52:03, Uri Blass wrote: >On September 26, 2001 at 17:47:09, Robert Hyatt wrote: > >>On September 26, 2001 at 14:27:51, Gian-Carlo Pascutto wrote: >> >>>On September 26, 2001 at 14:17:49, Uri Blass wrote: >>> >>>>I guess that another possible easy way for beginners is random changes in the >>>>evaluation by not more than 5%. >>>>If you change the value of the pawn from 1 to +1.05 the program may have a good >>>>chance to play a different move and the game is not going to be the same. >>> >>>The problem of this is that it may make the search instable. >>> >>>Georg v. Zimmerman had a nice trick here: before starting the search, >>>change the piece-squares a bit randomly. It gives the same effect but >>>is much more stable. >>> >>>-- >>>GCP >> >> >>That also means you can't keep the transposition table scores from the previous >>search or you will have _lots_ of debugging fun. > >I think that it is better simply to do it not before every search but only >before every game so you do not have this problem. > >Uri OK.. That would work fine. I was thinking about changing them after every search. At the start of a game would work just fine... although in my case, position learning would need to be turned off as that carries scores from game to game which would confuse things.
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.