Author: rasjid chan
Date: 16:58:00 04/05/04
Go up one level in this thread
On April 05, 2004 at 19:25:33, Sune Fischer wrote: If I am not wrong, this is what I am doing and my hashing SEEMS working ok. The PROBLEM is I make new discoveries every now and then. I understand hash table may have less bugs if we follow fail-hard ie NEVER FAIL OUTSIDE ALPHA OR BETA AND NEVER HASH OUTSIDE ALPHA BETA. But I have used fail soft without too much thought but simply because having a highest LB (> beta) or the lowest UB ( < alpha) seems appealing and trendy... I set a test question to see if most have the same answer. Assume we use fail soft and we use one TT for all nodes, hashing depth = 0 for QS, other things as usual, if there is anything that may be assumed. We are in QS and not in-check and we have a few valid moves searched and let :- x = highest score of all the moves searched x < alpha We now want to hash and return from QS. Poser :- How do you hash this node and return. Thanks Rasjid Rasjid >On April 05, 2004 at 19:10:59, rasjid chan wrote: > >>On April 05, 2004 at 18:58:57, Andrew Wagner wrote: >> >>This is what I learn, maybe from a cursory read of crafty's commentary >>(not source) and off I go... for Everest. It seems others are saying >>MUCH MORE. >> >>Rasjid > >You need the following information in the hash: >depth - to the leaves (called draft) >score - the value returned from the node >flag - was the node (hence score) a beta, alpha or pv(exact) node? > >Now when you probe from the search you have: >current_depth - to the leaves >alpha - lower window value >beta - upper window value > >So you can see there are a lot of combinations, each combination >must be handled. >For instance if the hash indicates: > >if ( >depth>=current_depth && // draft is usable >flag == BETA && // it was a fail-high (lower bound) >score >= beta) // the score is at least higher than beta, we fail-high >return beta; > >-S.
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.