Author: Richard A. Fowell (fowell@netcom.com)
Date: 12:07:01 03/08/98
Go up one level in this thread
On March 08, 1998 at 11:43:09, Aaron wrote:
>excuse me if this topic has being covered.
>
>
>It seems to me that chessmaster has this unique ability for users to
>tweak the program's chess "style" .Has anyone experimented and found a
>good setting for this program?
I've seen such posted from time to time ... here's one such,
but I have no idea how good it truly is:
---------------------------------------------------------------
Hello,
If you have CM 5500 and you wish make stronger programm.Make so:
Chessmaster 5500 with modified Standard Chessmaster personality - the
so-called "Walter-Pilz-settings" (Att./Def. -4, Select. 10, Mat.Pos. 4,
Mob. 103, King Saf. 170,Pawn Weakn. 125) plus my special piece values
(Queen 10, Rook 5.5, Bishop 3.5, Knight 3.3 - similar to MChess 7).
Increased hashtable size. Standard opening book.
!!! Supposedly stronger than CM 5500 !!!
---------------------------------------------------------------
>
>Also are there other programs that allow tweaking the style in such a
>manner? I know some allow limited styles..Solid,Agressive,active etc
I believe that Gromit might.
>Personaly I think all chess programs should allow this option such that
>even non-programmers can help to uncover good styles or Evaultion
>tables.. by playing Chess style 1 vers Chess style 2..
>
>Anyone done a experiment.. say play 10 games of Chessmaster 5000/5500
>using various personalities against another top notch computer say Fritz
>5.0/Rebel 9, virtual chess2 ,.
This is somewhat harder - unless CM supports Auto232 (it may, but it
seems unlikely), this will be incredibly tedious. The CM facility for
pitting CM personalities against each other is a much easier route for
tuning. Unfortunately, given the stochastic nature of these things, even
the limited goal of finding settings that have an edge on the standard
settings in this arena using blitz time control would take months or
perhaps years to do right, given that getting a 95% confidence level on
any given perturbation to 60 Elo takes 100 games (see the SSDF quoted
tolerances, for example) This suggests that any given perturbation
should take 694 machine-days to evaluate to 6 Elo at 95% confidence.
Given the number of settings available to perturb, this is fairly
daunting even if interaction of the various settings were weak. (note -
this time will be much reduced to the extent that the engine is
deterministic).
This actually might make an interesting Internet cooperative project
along the lines of the code-cracking projects. It would go something
like this:
1) When you log on to Chessmaster online you would be asked if you wish
to participate in improving CM5500 strength using "spare cycles" on your
machine.
2) If you assent, a test engine is downloaded, and recieves it's first
assignment from the central test coordination program on CM online.
3) The test program runs the match in the background until the test
segment is done.
4) The next time you log in to CM online, the test program reports its
results and receives its next assignment.
This is essentially the way that the cooperative code-cracking projects
work.
Note that testing can be facilitated here - since the coordinator knows
the probability for each CM5500 line, the opening lines can be tested at
their
known probability rather than stochastically. Also, the positions where
the
CM 5000 book is left can be played in "duplicate" fashion - each of the
two versions being compared having the same number of Blacks as Whites.
Note that, at the end of all this, all you know is:
"This set of settings can be expected to do well against the CM5000
stock
personality at blitz in CM5000 "tournament" self-play mode".
>
>Alternatively, use position tests would a difference personality allow
>the computer to solve certain tests faster or even allow it to solve
>test that it usually couldn't?
I think we can count on that.
>
>Would there be a great difference in results? I would think probably
>not. on the other hand if yes, so would rating lists have to take into
>account such differences?
One could certainly make a "LCTII" or "BT2630" chessmaster personality.
It would be even easier if CM5000 had a "batch test" facility - does it?
>
>More to the point , does anyone know what exactly we are changing in the
>evlulation tables when we alter the specfic variables..say Attacker vers
>Defender etc..would the progammer of Chessmaster 5000/5500(Wchess
>engine? ) like to comment?
? I thought CM used "The King"? Did they change this?
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.