Author: Carmelo Calzerano
Date: 13:48:36 06/23/00
Go up one level in this thread
On June 23, 2000 at 16:02:23, Adrien Regimbald wrote: >Hi, > >>This is a known hashing problem. Both repetition _and_ 50 move rule draws are >>are problem and cause the same thing to happen. You can either (a) ignore it >>or (b) include path information in the hash signature, which will solve the >>problem, but basically wipe out hash hits. > > >Hmm.. I obviously can't do (a) if I'm interested in getting decent results .. in >a 16 game match with SSEChess, Faile gave away 3 full points in easily winning >positions by allowing SSEChess to draw by repetition on SSEChess's move. Most >of these were cases where the game was an ending where Faile has an easy win as >soon as its king finds some cover so it can safely push the extra pawns it had. > >(b) is very undesirable, as the benefits of hash tables are tremendous, and I'd >REALLY hate to throw away all of that work after the frustrations I went through >to get everything more or less in order! :P > >Surely there must be a better way? > >Any ideas would be appreciated, What about checking for such kind of repetitions just at the root of the tree? I mean, for every legal move, check if opponent has a chance to get a threefold repetition and adjust the move score accordingly. This will not completely solve the problem, because you'll still miss some draw detection during search, but I think these cases will not occur so often: you will just repeat the position two times, and if possible avoid the third. Not perfect anyway, but fast and cheap: at least, I think this could be a good tradeoff. Best wishes, Carmelo Calzerano
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.