Author: Christophe Theron
Date: 09:42:37 12/02/98
Go up one level in this thread
On December 02, 1998 at 03:22:10, blass uri wrote: > >On December 02, 1998 at 01:45:10, Christophe Theron wrote: > >>On December 01, 1998 at 20:36:09, Kim Hvarre wrote: >> >>>Hello Ed, >>> >>>Exiting project You and Chistophe have going with The Tiger. In this relation >>>just one question, quoting Your homepage: >>> >>>"One of CHESS-TIGER's search algorithms has been implemented in REBEL and is >>>further improved by Ed Schröder. The CHESS-TIGER idea plus Ed's own improvements >>>have speed up REBEL with a factor of 2-3. The first results are very promising. >>>At the playing level of 1:00 per move REBEL scores 47½-31½ against various >>>strong computer opponents. >>> >>>For the moment the conclusion is Christophe's idea has given REBEL an elo jump >>>of at least 30 elo points but more likely the gain in playing strength is 50 elo >>>points or even more. The basic idea is im-plemented in the third party engines >>>of:" >>> >>>The question: The large jump in speed is this on behalf of something else >>>(knowledge, whatever) or is it just a "simple trick"/algoritme/?, You and >>>perhaps others have overseen? >>>If so, it's a rather amusing and a nice perspective in respect to the promising >>>future of search. >>> >>>Regards, >>> >>>kim >> >>Maybe Ed will answer too, but I can give you some info. >> >>As Ed and I have suspected right from the begining of the project, Rebel and >>Tiger are very very different programs. >> >>It appears that we use different algorithms to achieve a relatively close level >>of play. >> >>Some of these algorithms are incompatible with each other by nature. Maybe we >>will be able to get something by exchanging them, but it is not sure. >> >>But there are other algorithms that are compatible in both programs. And this is >>very interesting. Ed has found ideas that I have not found, and the opposite is >>true also. In the case Ed describes, one of Tiger's selection algorithms that >>did not exist in Rebel has been successfully implemented in Rebel. > >Does this selection algorithm speeds up Rebel by a constant factor or maybe the >factor is a function of the time per move? > >> >>This algorithm is not really a programmer's trick. It is an idea taking >>advantage of the nature of the chess game. I cannot say more, but I would rather >>describe this as being knowledge, because it would certainly not work with other >>games. > >Does it work in all the cases in chess or maybe it only works practically in >99.99% of the cases. Rebel scores much better in matches. I would say it works much more often than it fails. >I know that there are positions that Fritz5 can never solve because of the null >move(it does not analyze moves with no threat in the middle game). > >These positions are rare so practically it is not a problem but it can be a >problem in solving studies > >Are there problems with some rare positions with the faster Rebel? Ed should answer this. I my case, the new algorithm does not produce new "holes". In the worst case a combinational key move can be found a little later, and that's all. In most positions it is a clear win. Christophe
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.