Author: Dan Newman
Date: 02:37:38 08/13/00
Go up one level in this thread
On August 12, 2000 at 16:22:37, Adrien Regimbald wrote: >Hello, > >>You'd be surprised how little difference it makes to the overall search speed in >>normal positions. ColChess spends a negligible fraction of its time testing for >>checkmate, and I really don't think that it affects its ELO negatively at all. >>Far from it, in those positions where it is advantageous it often helps ColChess >>to win from equal (or worse) positions against stronger programs. >> >>Remember that a doubling in search speed probably only improves ELO by 60-70 >>points. I'm talking about a speed decrease of probably less than 1%, and the >>occasional benefits are quite substantial. > > >You're completely missing the point here. If you want to guarantee that you >have correctly ascertained if there is a mate in <x> from any given position, >you need to do a brute force search up to depth <x>, possibly depth <x+1> >depending on how you do mate detection. > >If you aren't talking about doing brute force searches, then you aren't >guaranteeing you see all the short mates, in fact you can be missing mates all >over the place. > >If you are talking about doing brute force searches, then you're an absolute >genuis, as nobody in the history of computer chess has managed to do a brute >force search at only 1% slower than doing the basic alpha-beta, or whatever >search method you use! > > >Regards, >Adrien. This sounds slightly incorrect to me. What I (and most others) call "brute force" search *is* alpha-beta. If you mean that you must do a search without alpha-beta to guarantee no missed mates, this is incorrect. Alpha-beta misses nothing that pure mini-max finds. If you do an alpha-beta search out to depth <x> without any reductions, you are guaranteed to find all the mates within that range (to depth <x-1> if you need the extra ply for mate detection). Now if you start doing forward pruning (Null move, etc.) or you do various reductions and so forth, you might begin to miss mates within the horizon. But alpha-beta pruning is a sort of freebe since you get the same result as pure mini-max. -Dan.
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.