Author: Dann Corbit
Date: 17:52:13 07/06/99
Go up one level in this thread
On July 06, 1999 at 19:33:08, Paul Richards wrote: >On July 06, 1999 at 13:53:22, Robert Hyatt wrote: > > >>Don't forget the 'math'. 99.9999999999999% of opening theory does _not_ follow >>moves like 1. a3 or 1. h3. So if a GM starts out like that, then the computer >>is going to be on its own. And there is no way to make a book wide enough to >>cover that... > >Forgive my ignorance, but why not? I was under the impression that Mr. Corbit's >project was doing just that. In fact it seems conceptually rather easy to >negate this aspect of anti-computer chess. Gather whatever theoretical and >statistical data exists on such offbeat openings, and in the absence of such >data crunch the resulting positions at suitably long time controls and build a >book response. All you need is a big enough book and time to do the analysis. >The book storage is just a matter of hard drive space, and the time to do the >analysis would be easy with a large group of volunteers. I'm happy to let Mr. >Corbit correct me if I'm wrong, but in a nutshell I believe this is already >being done. There are an awful lot of decent looking lines out to (say) 20 ply. One phase of the C.A.P. project, which I call "Bulldozer" will brute force the first 5 plies. Then "encouraging" lines will be explored from there. Already, by 5 plies, the number of possibilities is surprisingly large. I am not sure exactly how large it is, since I have not calculated the "fake" e.p. reductions yet (those records tagged as e.p. but the e.p. can't have any effect, so just mask it...) >Having a team to do analysis also lends itself to some pretty heavy-duty >learning, in that rather than just flagging moves as bad, if the program gets >itself into any bad situations or gets taken out of book early, you can produce >alternate lines or fill in any missing areas rather quickly. Such a book would >just keep getting better with time, and I think in time would effectively negate >this particular anti-computer tactic. Instead of getting the program out of >book the GM might find himself either in an unenviably silly position or else >transposed into a normal opening with a bad move order or a lost tempo. The C.A.P. project will remain ongoing precisely because the problem is so large. And each new level of detail will reduce the probability of serious difficulties -- but not eliminate it.
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.