Author: Peter Fendrich
Date: 16:20:54 10/10/02
Go up one level in this thread
Well, I do it myself but have never tested it because there is no extra cost involved and it can't at least be worse. Before starting a rootmove search I fill the saved pv into the hash table and that goes for all root moves. Not just the current best one. Peter On October 10, 2002 at 18:22:16, Dieter Buerssner wrote: >On October 10, 2002 at 17:51:17, Peter Fendrich wrote: > >>On October 10, 2002 at 17:17:43, Dieter Buerssner wrote: > >>>You are correct, and the example I gave was bad. Other posters explained it >>>better. To my excuse, I might say: For my engine the example was mor or less >>>correct, because I also update "PVs" (which may be be called refutation >>>variation) when outside of the alpha-beta window (so for a fail low node, >>>whatever the fail soft alpha beta loop reported as best move, for fail high >>>node, the move that returns a score >= beta). I use this "refutation PV" to help >>>moveordering, for example in the case where a reasearch from zero window search >>>is needed. >> >>Will this really help? > >In my experience, it did help a bit. > >>The fail low node will not produce anything useful and the fail high node move >>will be kept in the hash table so when researching it will be there anyway. wont > >Yes (but I think even the fail low, can give a bit of info, when using fail soft >search at least.). But for my engine, there is often the situation (in games >with longer TC) that the hash tables are overcrowded, and some of the essential >info can be overwritten. The "refutation-variations" should give a persistent >source for the move ordering in this case. Note, that this idea is from old days >with a 396SX-16 and little memory (I started with 1MB). Perhaps, I didn't have a >hash table at that time (I cannot remember). I certainly tested it later with >HTs, and it still gave slightly smaller tree sizes in the situations I tried. > >Regards, >Dieter
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.