Author: Robert Hyatt
Date: 18:14:13 02/03/98
Go up one level in this thread
On February 03, 1998 at 17:44:40, Ed Parry wrote: >Hi Guys! > >I have seen LAZY EVAL referenced here and RGCC a few times and am not >sure what it is. Could someone give a brief explanation please? Here's the basic idea. Suppose the values for alpha and beta are -10 and +10 when you reach Evaluate(). The first approximation to the final score is always just a simple "count the material" approach. Suppose the material score comes out -3.00, white is a whole piece down. Can your Evaluate() procedure come up with positional scores of +3, to bring the score back inside the alpha/beta window? If not, there's no need in going any further, just return alpha and save the time that you'd spend evaluating this node. If you can produce such large scores, first do the evaluation things that *add* points. If, after doing all the adding, the score is still < alpha, what's the point of doing the negative things since that will only drop it further below beta. "lazy eval" is the name coined by being "lazy" and not doing *anything* in the eval that doesn't seem to be needed. IE after you do all your plus stuff, maybe the score is now > alpha... then you start doing the negative things... you can quit the instant the score drops back below alpha, since that's enough to cause a cutoff... it doesn't matter how much lower the score can go... > >I also FINALLY got a working program going thanks to the excessive help >of James Long (Tristram author). I have a long way to go and am finally >able to understand better the sources I have - crafty, tristram, >minimaxe, TSCP, etc. that I have. > >I'd just like to thank JL openly for his patience, help and suggestions. > >I'll most likely be asking more advanced questions as I progress, but I >have a huge list of stuff to do now just to get the program decent and >stablized. > >Anyone have an suggestions? I basically have SEARCH working finally >(based on Tom K's NEGAMAX C source) and my evaluateposition() function >is based ONLY on material and a simple center table. THis needs alot of >work obviously. > >Also, I need to add in some form of move scoring in the move generation >phase so I can get some form of order to my search. Any suggestions >beyond MVV/LVA? use SEE is my first suggestion, so you can look at captures that win the most material first (MVV/LVA will look at QxR before NxP, even though the R is defended and the P is not) while letting you look at the lemon captures later... then work on move ordering... you get faster without changing the scores or moves, which is good... > >For those of you running sites re: chess porgramming - KEEP IT UP! I am >learning quite a it just by reading the overview of your rpograms. Same >with thsoe releasing openly your chess programs. I LOVE reading the tech >aspects how-to's - what I do's, etc. > >Also to those releasing source - make sure you keep up the comments! >Tristram and Crafty are two of the best commented sources I have seen. > >Thanks to all that helped me in the past (BH, TK, DT, JL, BM, ...) > >FTR, I searched Springer-Verlag ( http://www.springer.ne ) for CHESS >SKILL IN MAN AND MACHINE and they claim they have two editions >available. A 1983 edition ($53 USD) and a 1984 ($30 USD) edition. Anyone >know the diffs? I asked them, and they've yet to reply (been about a >week). > >Off I run - we are raining hard in Southern California - Ed new edition is better, it has more chapters on a few more recent topics. Both cover chess 4.x of course...
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.