Author: Robert Hyatt
Date: 07:13:48 12/26/01
Go up one level in this thread
On December 26, 2001 at 04:52:35, Sune Fischer wrote: >On December 25, 2001 at 23:56:56, Robert Hyatt wrote: > >>On December 25, 2001 at 20:21:49, Sune Fischer wrote: >> >>>I'm sitting here with an implementation problem, I can think of a few ways to >>>solve it, but perhaps there is an even better one. >>> >>>The problem is I can't always get the best move in a position when searching, I >>>need a beta cutoff or an improvement in the alpha value. So it isn't save for me >>>to assume there is killer in all recorded hash entries, and if I only >>>recordhash() when I _can_ put a killer in, then I miss out on some scores. >>> >>>So I was wondering what solution you guys have found? >>> >>>-S. >> >> >>If I understand your question, there are three kinds of positions where you >>can store a hash entry. PV entry where the score is _EXACT_. You should >>have a best move to store hre. >> >>LOWER entry which is AKA a fail-high position. A move caused this fail high >>and it should be stored in the hash as best. >> >>UPPER entry which is AKA a fail-low position. There is _no_ best move in this >>position. I simply store a 0 for such cases... > >Yeah that's what I mean, the first thing i do in the AB search (I'm not using >PVS yet) is to probe, and if succesful I put the killer into the pv and return >with the score. >At the end of the search I expect there to be a move in the pv, namely the best >move the search could find. What if the entire search is returned from a probe >hit, one that did not have a move stored, then the pv is empty! >I hasn't happend to me yet because most entries do have a killer, but it could >happen I guess. > >-S. It will _bury_ you if you do aspiration search at the root. IE you don't start with +/- infinity, but some narrower window. If your window is too "high" you will fail low at the root and you won't have any move from the hash table. But in my case this isn't a problem since I don't probe the hash table at the root anyway...
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.