Author: Dave Gomboc
Date: 21:10:01 12/07/00
Go up one level in this thread
On December 07, 2000 at 16:22:49, Bo Persson wrote: >On December 07, 2000 at 05:34:08, Leen Ammeraal wrote: > >>On December 07, 2000 at 04:23:48, Andrew Williams wrote: >> >[...} >>> >>>My program uses MTD(f). It stores an upper bound and a lower bound >>>for each position in the table (these are set to +HUGE and -HUGE in >>>empty entries). It also stores a draft for each bound in the same >>>hash entry (these are both set to -100 or something in an empty record). >>> >>>Cheers >>> >>>Andrew >> >> >>Andrew, thanks. >>I had just come to the same conclusion: storing >>two bounds requires storing a depth (or draft) for >>each, and I have already been implementing this >>and am now testing it. >>I still fail to see why this should only be useful >>for MTD(f), not for alpha/beta or negascout. >> >>Leen > > >Yes, I have been doing like this since I started trying out MTD(f), about a year >ago. Having both bounds is useful for negascout as well (accidentally >tried that in an intermediate program version...), but absolutely *essential* >to get any performance from MTD(f) re-searches, as it might swing back and forth >at the end of the search (closing in on the final score with a slight >overshoot in each direction). > >Contrary to what Bob says in another part of the thread, I think that storing >both bounds in the same entry actually *saves* space, as you otherwise would >have empty best moves in the upperbound nodes. > > >Bo Persson >bop@malmo.mail.telia.com FWIW, in my Awari program I did it like you suggest: one best move, two bounds, two depths in one hash entry. Dave
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.