Author: Bo Persson
Date: 12:47:55 03/03/04
Go up one level in this thread
On March 02, 2004 at 18:07:18, Volker Böhm wrote: >On March 02, 2004 at 15:05:33, Robert Hyatt wrote: > >>On March 02, 2004 at 14:50:43, Volker Böhm wrote: >> >>>Hi, >>> >>>in the post "Simple rook sac" I´ve found the following crafty analysis result: >>> >>> time=8:36 cpu=395% mat=1 n=1370692317 fh=92% nps=2.66M >>> ext-> chk=58001759 cap=3294306 pp=293332 1rep=5509951 mate=376022 >>> predicted=0 nodes=1370692317 evals=299470745 >>> endgame tablebase-> probes=0 hits=0 >>> SMP-> split=2056 stop=369 data=9/32 cpu=33:59 elap=8:36 >>> >>>Crafty got 300M evals and 1370M nodes. >>>My Question: >>> >>>Why has crafty got that much less evals than nodes? What does crafty count as >>>node and when does crafty don´t need to eval. Non leaf-nodes, hash-hits with >>>right depth and value range, anything else? >> >>all of those, plus "lazy eval" cutoffs that avoid doing a full eval... > >When do you use lazy eval? If (material + bonus < alpha) || (material - bonus > >beta)? how big is bonus? Do you count passed pawns in "material"? >For me passed-pawns + king-attack + pawn-shield could get up to 6 pawns >"positional" value. Is your bonus much less than your theroretical positional >value? > You can also be "a bit lazy". Start with the easy and/or potentially high valued evaluation terms, and give up when you realize that you will never reach alpha. Maybe a smaller gain each time, but it can happen more often. Bo Persson
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.