Author: Robert Hyatt
Date: 18:36:31 09/26/01
Go up one level in this thread
On September 26, 2001 at 20:55:33, Miguel A. Ballicora wrote: >On September 26, 2001 at 19:18:18, Gian-Carlo Pascutto wrote: > >>On September 26, 2001 at 19:03:12, Miguel A. Ballicora wrote: >> >>>One of the big problems of alpha-beta is that the machine is quite >>>stupid to determine how to traverse the tree and how to prune (pruning is not >>>part of the pure alpha-beta actually). >> >>Hmm? >> >>I'd say beta cutoffs are a perfect example of pruning. >> >>Not sure what you mean by 'pure alpha-beta' though > >No null-move, not futility-cuttofs, no forward pruning of any kind, >no extensions etc. Any sort of trick that affect the shape of the tree. >i.e. speculative tricks. > >>Making something faster (in terms of nodes) than alpha-beta with >>perfect move ordering is simply impossible unless you start doing >>tricks with the hashtable (and try to reach the real minimal graph >>instead of the minimal tree). > >You do it actually with null-move, so it is not impossible. I don't like that statement. Null-move does _not_ make alpha/beta faster. It makes it both faster _and_ more error-prone. There is a difference between "faster but same result" and "faster but with more errors in result." > >Besides, why are you assuming that the programs nowadays have a perfect >move ordering? in fact, it is good but still there is a lot of garbage >that can be thrown out, even if the ratio of cutoffs is >90% since that >10 % is propagated across the tree. First I would say that perfect move ordering is impossible, by definition. It is a chicken / egg problem. You can't know you have perfect move ordering until you do the search. And once you have done the search, it is too late to correct any imperfect move ordering. Perfect move ordering is the "Holy Grail" of computer chess. Once we have that, chess is solved, because we simply play the first move this perfect ordering algorithm produces and we don't need to search at all... > >Regards, >Miguel > > >>If alphabeta is the wrong answer, you were asking the wrong question. >> >>-- >>GCP
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.