Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: PVS researches and best lines

Author: Gian-Carlo Pascutto

Date: 08:26:13 12/11/00

Go up one level in this thread


On December 11, 2000 at 09:55:34, Rémi Coulom wrote:

>I am not sure I understand all you wrote, but I see a small problem in using
>-value+1 instead of -value: it should not change anything in practice if you >are using a hash table.

I do not use a hash table. I am using a stripped-down search that does
not do nullmove or any kind of pruning either. Just PVS. My program does
have all those features, but I've killed them out in order to fix
problems with basic functionality, like e.g. best line display.

I've got it working to the point where I can use aspiration search and
PVS without losing any PV's.

I'm now running into the problem that I get fail-low's after failing
high on the same score, if I enable nullmove and hashtables. I know
this is possible if you use forward pruning and pondering together,
but I'm not pondering. Also, it still happens if I disable the usage
of the scores and bounds from the hashtable, and only use the suggested
best move. I do not understand why I am getting this behaviour (different
scores) if I only change move ordering.

>I do not know of an easy way to get the full PV all the time. Maybe those who
>use MTD(f) have thought about it. I think that if there is a way, then it must
>use the hash table because of the problem I have mentionned above. I do not >know how, though. This does not look like an easy problem.

Crafty, EXChess, ... seem to always get the full PV. If I use the
above trick I do too, but I cannot use the bound from the hash to
tighten the window or I lose it again.

--
GCP



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.