Author: Robert Hyatt
Date: 10:12:33 09/02/02
Go up one level in this thread
On September 02, 2002 at 06:49:34, Tony Werten wrote: >On September 01, 2002 at 23:47:17, Robert Hyatt wrote: > >>On September 01, 2002 at 13:44:45, Gian-Carlo Pascutto wrote: >> >>>On September 01, 2002 at 13:26:44, Robert Hyatt wrote: >>> >>> >>>>>pawn=32 in fritz seemingly. that's all you need to know to consider >>>>>it works for it. >>>> >>>>What does that do? I have seen large positional scores out of fritz, >>>>which suggests (to me) that mtd(f) could cause some problems... >>> >>>That means that Fritz has a low evaluation granulatiry, which keeps the >>>number of MTD passes to a minimum. There's some disagreement here between >>>Vincent and the rest of the world exactly how much this matters, but it >>>does matter for sure. >>> >>>Positional scores have nothing to do with this, I don't know where you >>>got that. >>> >>>-- >>>GCP >> >>I have no idea what you are talking about above. The problem that mtd(f) >>encounters is with an eval that fluctuates significantly iteration to iteration. >>That fluctuation is _not_ an issue of pawn values. It is an issue of >>positional scores. You could have pawn = 10000 if your positional scores >>don't vary much. But if they vary by more than a pawn, you will have trouble >>no matter what... >> >>reducing the score range by a factor of 3 will help _some_. But only _some_. >> >>But when a program can produce scores with the positional component well over >>two pawns, I don't think the actual pawn value has much effect on how mtd(f) >>performs... the constantly shifting eval is going to cause lots of re-searches, >>from experience.,.. > >If the score difference is .25 of a pawn then with pawn=32 you need 8 mtd >researches, with pawn=1000 you need 250. Or am I mising something ? > >Tony Yes. but suppose the positional scores are +/= three pawns, as I have seen out of fritz? And of course nobody would do 250 searches, there are better algorithms since fail-soft gives you a better bound for the next attempt, rather than simply adding one if you fail high...
This page took 0.01 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.