Author: Robert Hyatt
Date: 17:11:27 11/09/97
Go up one level in this thread
On November 09, 1997 at 18:00:30, Ed Schröder wrote: >>Posted by Robert Sherman on November 09, 1997 at 12:37:35: > >>This post is not meant to demean Rebel 9 in any way. Rebel 9 is, in my >>opinion, a very strong program with many nice features that makes it a >>significant upgrade over Rebel 8. Although Rebel 8 and Rebel 9 are very >> strong, I have uncovered a glitch in respect to the amount of time the >>programs spend on taking the piece on certain sacrifices. The following >>game was played on a "obsolete Pentium 100", so it may not be applicable >> to the faster machines. I did send this game to Ed Schroder when he >>was producing Rebel 9. On a fixed level Rebel 9, does see the impending >>mate 15 seconds later then it does when playing at a 40/2 tournament >>level. This may mean that Mr. Schroder was testing Rebel 9 on faster >>machines and didn't believe that the need to make changes to the way >>Rebel 9 allocates time would make any sense. My belief is that when a >>sacrifice is played, the program should realize that move is very >>critical for the game and spend a long time analyzing whether the sac is >>sound or not. Anyway, on the fixed time level, Rebel 9 finds a >>refutation the second fastest of programs I have tested, second only to >>Genius. Also, shareware programs never reject taking the knight bait >>while taking deep searches. While all commercial programs find the >>looming mate (except Rebel because of the timing glitch) and make the >>appropriate move. So this game seems to separate the "men from the >>boys." I have not tested this game on Crafty or some commercial >>programs (Hiarcs 6, CSTal, Shredder), so I would be interested to see >>how they do. Without further ado, here is the game: (Sorry I didn't have >>time to put it in PGN) > >I am aware of this for years. Rebel is programmed for fast recaptures. >This algorithm of course has some negative side effects shown in the >below game which fortunately are big exceptions. > >Still I prefer to leave it the way it is now because removing the >algorithm would mean Rebel will think 3-5 minutes again on simple >recaptures. I received too many complaints about that in the past. > >Thus so now and then you are lucky :) > >- Ed Schroder - > > there's two solutions here: 1. reply quickly *only* on recaptures, rather than on any capture that seems to be better than any of the other moves by a significant amount. I do this. 2. Don't play O-O into that attack in the first place. Crafty concluded that O-O was totally busted, after a minute on a P5, more like 30 seconds on the P6/200... That's a difficult problem however, when you have to choose: (a) castle short *now*. (b) castle long, but it takes a few moves to prepare; and (c) don't castle period. I've spent a good bit of time trying to make this work, which turns out pretty well most of the time... > >>White : Robert Sherman Black: Rebel 9 > >>1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. Nc3 Nc6 5. Nd5 Nxe4 6. d3 >> Nxf2 7. Qe2 Nxh1 8. Ng5 O-O 9. Nxc7 Qxc7 10. Qh5 h6 11. Bxf7 >>Kh8 12. Qg6 Qa5 13. c3 Qxc3 14. bc Bf2 15. Kd1 Ne7 16. Qh7#
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.