Author: Robert Hyatt
Date: 15:46:43 01/19/04
Go up one level in this thread
On January 19, 2004 at 13:34:56, Sune Fischer wrote: >On January 19, 2004 at 12:11:30, Tord Romstad wrote: > >>On January 19, 2004 at 11:43:21, Sune Fischer wrote: >> >>>A while back I ran lots of test positions and compared the PVs between the two >>>methods. >>>What I generally saw was that they were identical, only the hash extracted PV >>>usually were a few moves longer. >> >>I extract all my PVs from the hash table. I am not sure I agree with >>your observation that the hash extracted PVs are usually longer, but I >>agree that they are *sometimes* longer. > >It makes some sense though, if Crafty gets a hash hit on a PV node the PV will >get cut right there, while ours will have no problems being extracted into the >part that created the cutoff. Or not. If you look back thru the comments in main.c, I used to do this to extend the PV. IE on a hash hit, I would probe forward to pick up other moves as well. And I found nonsense on occasion and decided to stop... > >The only way our PV's can become shorter is if a needed entry actually gets >overwritten, the probability of that is not so great in a short search/big hash. >Unfortunately it is definitely going to happen in a long search. > >>Today Leo helped me discover an obscure bug related to this. The >>maximum search depth in Gothmog is 60 plies, and my PV[] array has >>60 entries. The fact that extracting the PV from the hash table >>could cause a PV of length greater than 60 plies had never passed >>my mind until Leo sent me a debug file from a game where Gothmog >>crashed directly after printing out a 61-move PV. :-) > >I helped me to fix a few bugs too, some hash bugs are seen instantly this way :) > >I have a PV class now for extracting the PV, my pv.add(move) does a bounds check >before adding the move. >This is one place where bounds checking is called for, doing it only at the root >is 'free' anyway :) > >-S. >>Tord
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.