Author: Larry Griffiths
Date: 15:49:42 09/02/00
Go up one level in this thread
On September 02, 2000 at 17:24:24, Carlos del Cacho wrote: >On September 02, 2000 at 16:49:11, Larry Griffiths wrote: > >>I am adding "EXACT" hashtable entries at each ply when there are no cutoffs. >> > >This could be your mistake. You should make a difference between a PV node (a < >value < b) and a fail low (when all your moves yield a value below alpha). You >can't flag a fail low as an exact value value because it isn't such. Your >opponent is failing high and maybe the best move wins a rook when he just needs >to capture a pawn to get a beta cutoff. In those cases, all you know is that the >value of the subtree searched is equal or less than what you got. If I >understood you correctly this is the source of your problems. > >So, this is the typical hash scheme (after checking that TT depth >= actual >depth) : > >- fail high => store value as a lower bound on the true score of the subtree >from that node. If you get to one of these and the retrieved value from the >table is > beta, then fail high without further search (it can only go better >and we already know it's over beta). > >- exact => just return the value from the TT. > >- fail low => this is an upper bound, so if it you get here and that value is < >alpha you just fail low since you know it just can get worse. > > >carlos Thanks Carlos, I think I get the gist of what you are saying. It looks like my definition of EXACT is way off. All this lower/upper bound stuff makes my head spin. I will save this post and see if something sinks in over the next few months... Larry.
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.