Author: Ed Schröder
Date: 09:42:34 06/22/98
Go up one level in this thread
>Posted by Don Dailey on June 22, 1998 at 11:03:26: >>Comments... >>There are some doubts on the Mchess results. First in the very first match >>between Genius and Mchess a line by Mchess is played that isn't active in >>the Mchess book. The line is 1.d4 d5 2.c4 dxc4. I don't know how this >>could have happened. >>Next I found out that Mchess on CCL does not learn from its previous >>games. From the 80 games Mchess played sofar on CCL it shows up that >>there are 10 exact double games. With exact I mean 100% identical games >>from move one to the last move. These 10 double games are not in favor >>of Mchess as its score is ½ - 9½. >>This case is under investigation now and Marty and I have some emails >>going to find what is the reason and how things can be improved. >>If there is un-justice done to Mchess all Mchess matches will be >>replayed. >>Mchess 7.1 on CCL runs as follows... >>. Hash table 60 Mb >>. Book learner enabled >>. Position learner enabled >>. Level is 40/120 20/60 20/60 >>. Table bases enabled (600 Mb on hard disk) >>For the moment my (premature?) conclusion is that Mchess 7.1 doesn't >>learn due to the 40/2:00 level settings and that I should use another >>level instead but only Marty himself can tell. I will keep you posted. >>- Ed Schroder - >Ed, >If it turns out you are right about this, it will be a great >opportunity for us to see first hand how important this feature >is to Mchess. >However I would point out that the other programs (assuming they >are learners) still have an advantage since they will now be >"educated" when they re-play Mchess, but Mchess will have to start >from scratch. >I'm guessing this will not affect things very much since they will >be facing a new Mchess and have to re-learn themselves, but it's >likely to have some affect. >I hope you can make this as fair as possible by re-initializing >the learning feature on Mchess opponents. But this might influence >other games too with other opponents. >Can you comment on these issues since I am not sure how you deal >with them? If it turns out that you always re-initialize >the learners before it plays each opponent then my concerns are >a moot point (not an issue.) All programs started clean "out of the box" with the recommended settings from the manual which for most programs are the default settings after installation. I remember I only modified the hash tables sizes of Nimzo98 and Shredder 2.0 for their best settings to avoid the Windows swapping. For the other (DOS) programs this was (is) no issue since they are always started in pure DOS mode under W95 using F8. After that nothing has been touched anymore. So learn results are not deleted and will never be deleted. I don't know yet how to handle this case further since it is still unclear to me why Mchess did not learn. I checked SSDF games and Mchess learns on SSDF. Mchess apparently does not learn on CCL. Marty just told me that it isn't the level which I suspected. He gave me a few hints which I will check. I also checked the 1.d4 d5 2.c4 dxc4 case. I checked the original log files and found out that after the Genius-Mchess match was finished 2 games (from the 10) were aborted. As a result I had to play 2 more games and I simply mixed up the colors. So it wasn't Mchess who played the Queens Accepted but Genius. The error will be of course corrected. - Ed - >- Don
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.