Author: Miguel A. Ballicora
Date: 16:03:12 09/26/01
Go up one level in this thread
On September 26, 2001 at 17:56:29, Robert Hyatt wrote: >On September 26, 2001 at 16:53:56, Miguel A. Ballicora wrote: > >>On September 26, 2001 at 16:32:26, Gian-Carlo Pascutto wrote: >> >>>On September 26, 2001 at 16:20:25, Dann Corbit wrote: >>> >>>>But maybe Vincent has come up with a real stunner. >>> >>>We are all eagerly awaiting the proof that 1 + 1 > 2 >> >>On the other hand, I do not see a proof that can't be. >> >>In theory... why not? with iterative deepening you have 1+2+3+4+5+6+7+8 < 8 >> >>Regards, >>Miguel >> > >That is not the same thing. That is taking the very serial property of >alpha/beta and using it to advantage by iterating. Parallel searching doesn't >assist a serial property, it works against it. Of course, once you know how. If I told people loooong time ago that I would do iterative deepening and that will be faster people will say that is not possible (if they don't know the existence of tricks to improve move ordering such as hashtables etc). Easy to explain once you see it. 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). In fact, that is the reason why most of the tree is garbage. Most of the time it would be very useful to know if it is convenient to continue the path or quickly search other possibilities to see if I can terminate the search right now. Under alpha-beta, this kind of decisions are not possible without akward tricks. Nobody can guarantee that there is parallel algorithm that combined with a certain move ordering and pruning method can make alpha-beta better. Who knows? Breakthroughs are difficult to foresee! The fact that nobody imagined it in decades does not prove anything. Hard to believe? yes. One has to be skeptical but I do not think that we can say "impossible". You said that if there is a >2x speed up is because the sequential algorithm is bad. Well, maybe alpha-beta is already the problem. ;-) Regards, Miguel > > > >>> >>>-- >>>GCP
This page took 0.05 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.