Author: Mihaly Szalai
Date: 09:10:41 07/24/03
John, thank you for the explanation. I just thought of this because my CM rating is quite unreal this way - and since there's no position learning and randomness is zero, CM default pers. won't improve in this line. I don't think that adjusting the cfd value would be a great risk for the engine against a weaker player. If I'm a 2100 player most probably I will be the first to blunder, not the engine - even if I have a 0.30 - 0.70 edge at some point of the game. I used to play in the Hungarian Team Championship and when I met a player 100-200 Elo stronger than me, I offered a draw quite frequently but they never accepted it, thinking they are the better chessplayers and they will 'outplay' me sooner or later, so I had to fight until 2 kings - what is the right way in serious games IMO. As for the book lines: CM uses several opening books automatically (quite unique in the compchess world!). There should be books with no drawish, sharp lines against weaker players. And - believe it, or not - I don't want to know the engine parameters and opening lines when I play a serious (rated) game. (Otherwise I do!) Best to you Mihaly On July 23, 2003 at 22:24:16, John Merlino wrote: >On July 23, 2003 at 22:05:17, John Merlino wrote: > >>On July 23, 2003 at 22:02:32, John Merlino wrote: >> >>>On July 23, 2003 at 14:58:47, Mihaly Szalai wrote: >>> >>>>On July 23, 2003 at 13:58:43, John Merlino wrote: >>>> >>>>>On July 23, 2003 at 10:26:52, Mihaly Szalai wrote: >>>>> >>>>>>The CM 9000 default personality is rated 2789 on my Celeron 1.7 GHz >>>>>>machine. My Hungarian rating is 2127. Are we in the same class? No. >>>>>>It simply cannot handle this opening properly, unless it counts >>>>>>as success a draw with black (as I do against a comp with any colour). >>>>>> >>>>>> >>>>>>[Event "Chessmaster 9000 Rated Game"] >>>>>>[Site ""] >>>>>>[Date "2003.7.23"] >>>>>>[Round ""] >>>>>>[White "Szalai"] >>>>>>[Black "Chessmaster"] >>>>>>[TimeControl "3600+5"] >>>>>>[Result "1/2-1/2"] >>>>>> >>>>>>1.d4 d5 2.Nf3 Nf6 3.h3 e6 4.e3 Bd6 5.Bd3 c5 6.c3 O-O 7.O-O Nc6 8.Nbd2 >>>>>>Qc7 9.Re1 Bd7 10.a3 c4 11.Bc2 Ne7 12.e4 dxe4 13.Nxe4 Nxe4 14.Bxe4 f6 >>>>>>15.Bd2 Nd5 16.Qb1 f5 17.Bc2 Rac8 18.Ne5 Bxe5 19.Rxe5 Qd6 20.Qd1 Qb6 >>>>>>21.Rb1 Qd6 22.Ra1 Qb6 23.Rb1 Qd6 24.Ra1 1/2-1/2 >>>>> >>>>>Taking a look at the game, after 19.Rxe5, the Chessmaster personality feels that >>>>>White is slightly ahead (a 1-minute analysis on a P4-2.4 gives an eval of 0.31 >>>>>for White). So when you offered a draw by repetition with the White pieces, >>>>>Chessmaster said "thank you very much!" :-) >>>>> >>>>>jm >>>> >>>>Thank you, John. >>>> >>>>And what do you think of that: >>>> >>>>Chess programs should take into account the playing >>>>strenght of their opponents (at least in rated games), >>>>and choose their opening lines and playing style >>>>accordingly. >>>>Too hard task for a programmer? >>>> >>>>http://www.talkchess.com/forums/1/message.html?307709 >>>> >>>>Mihaly >>> >>>It is a good idea, I think. But, depending on how thoroughly you want to >>>implement it, it can add great levels of complexity (i.e. complications) to >>>something that is already difficult for an engine writer to control -- the >>>overall strength of the engine! >>> >>>Probably the easiest and simplest way to do this (and Johan and I did discuss >>>this during the development of CM9000 -- it got put in the "next version" pile), >>>is to ONLY adjust the engine's internal "Contempt for Draw" value depending on >>>the rating difference between the engine and its opponent. Of course, this would >>>be an option that the user could turn off, as it could result in MUCH weaker >>>overall play as the engine is now playing for a win instead of simply trying to >>>find the best move. >>> >>>To also try to add data to each move in the opening book that attempts to put a >>>value on the "drawishness" of each move could also prove to be a difficult task. >>> >>>jm >> >>I guess I could have added that there is one simple way to test this theory. >>Create a copy of the Chessmaster personality by editing ONLY the "Contempt for >>Draw" value to something like 2.0 or even higher. This could result in the >>engine playing the kind of game that makes it harder for humans to draw against, >>but probably easier to win against! >> >>jm > >And, just because I got curious, I decided to "analyze" the game using the >Fischer personality from CM9000, which has a contempt for draw value of 3.0. As >it turns out, it appears as though it would definitely attempt to avoid drawing >the game. > >After 21.Rb1, it would likely play 21...Rf7. >After 22.Ra1, it would likely play 22...Rc5. >After 23.Rb1, it would likely play either 23...a5 or 23...Rf7, depending on time >control and computer speed. > >jm
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.