Author: Jorge Pichard
Date: 14:25:14 05/26/00
Go up one level in this thread
On May 26, 2000 at 12:59:31, Christophe Theron wrote: >On May 25, 2000 at 17:09:37, blass uri wrote: > >>On May 25, 2000 at 16:43:20, Jorge Pichard wrote: >> >>>On May 25, 2000 at 15:29:43, Daniel Chancey wrote: >>> >>>>What are the settings on CMQueen+? >>>> >>>>Castle2000 >>> >>>Under the Game Bar after you select Set up a Personality >>> >>>Opening Book = CM6000 >>> >>>Style Positional Material >>> >>>A/D =5 M/P =10 Own Q = 9.9 >>>SOP =100 C/C =105 Opp Q = 9.7 >>>Rand..=0 Mob =120 Own R = 5.0 >>>MSD =20 K/S =100 Opp R = 4.9 >>>S/S =8 P/P =110 Own B = 3.2 >>>C/D =1.0 P/W =100 Opp B = 3.2 >>> Own K = 3.1 >>>TT =ON Opp K = 3.1 >>>DT =ON Own P = 1.0 >>> Opp P = 1.0 >> >>Can someone explain me what is the logic behind the unequal values of queen and >>rook. >> >>I do not believe that it is better than equal values without a clear evidence >>for it. >> >>I suspect that people got a wrong impression based on a small number of games to >>decide that the unequal value for queen and rook is right. >> >>I am not going to believe it without a logical reason. >>I can see a logical reason that it may be better than other programs if they do >>not know to play when there are queens on the board but I do not see a logical >>reason that it is going to be better in a match against the personality with >>equal values(9.8 for a queen and 4.95 for a rook) >> >>I also want to know if there is a test that prove with 95% confidence that the >>personality with the unequal values is better than the personality with the >>equal values. >> >>If there is no proof for it then I am not going to believe that the personality >>with the unequal values is better. >> >>Uri > > >I think the idea is that the program is less likely to trade queen and rooks if >it values its own Q and R more than the opponent's. > >It could be a good idea if the program is supposed to be superior in king >attacks for example. It is almost impossible to build a strong midgame king >attack without a queen (and a rook maybe), so you tell the program to keep its >attack material as long as possible. In most of the games that I obseved between Chess Tiger 12.Oe Vs CMQueen what you described in the previous paragraph is exactly the case in question; CMQueen+ Kept its Queen as long as it could thinking that it was better off, but as soon as Tiger obtained the best position with its Queen or Tiger Queen Got close to the CMQueen+ King then CMQueen+ immediately traded Queens. In the case of the rooks CMQueen+ always kept its rook more active then Chess Tiger 12.Oe but as soon as Tiger started to press CMQueen+ or place its rook on the seventh rank, CMQueen+ did whatever it took to trade its rooks as well. Now as far as its Bishops being rated a little bit higher than its Knights CMQueen+ avoided at any cost to trade its bishops for Tiger Knight, unless the position was such that in order to avoid being forked in the next 5 moves, then CMQueen+ immediately traded its bishops fot the knights. Jorge Pichard > >Or maybe CM6000 has problems in the endgame against Tiger, so staying in the >middlegame as long as possible helps... > >In general, asymetry in the evaluation is used when you make strong assumptions >on your opponents. > >For example, if you think you are stronger in pure tactics than your opponent, >you do whatever you can to open the position and place your pieces in wild >positions. Then you wait for the opponent's mistake. > >This is done with asymetry in the evaluation. A simple way to do it is to count >the open files and to value this as good for the program's side. So if you give >the program white in an open, balanced, position, it will value it at +0.5 for >example. And if you give it the black pieces, it will also value it at +0.5, >which is an obvious proof of asymetric evaluation. > >Of course, this will backfire if it happens that you are NOT stronger in >tactics. Asymetry is dangerous. It can work against a given opponent, and be a >disaster against others. > > > > 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.