Re: [Bug-gnubg] Simple playing strengths

[boomslang]

--- Jonathan Kinsey <address@hidden> wrote:

> The way you currently set the playing strength of gnubg isn't great.  For a 
> beginner all the options are meaningless, for a more experienced player the 
> settings don't easily relate to actual strength.
> 
> A simple idea is to replace the current player->settings with just a single 
> combo-list (with the 8 strengths).  The idea is that both the chequer and 
> cube settings will be set to this value.  An "advanced" button could then 
> pop up all the current settings.
> 
> BTW, does anyone now roughly what fibs ratings the current preset settings 
> map to?  Perhaps the settings could be tweaked slightly to broaden the 
> playing range and differenciate more between settings.  This is because I 
> often hear people (beginners) say "it's too hard/good even on the easiest 
> setting" and "it doesn't seem any different between setting [x] and [y]".
> 
> Any thoughts?
> 
> Jon


Hello, 


>From May 2005 until March 2006 I ran two GnuBG bots with various settings on
FIBS. They played in total about 67000 one point matches. For each setting the
FIBS rating is estimated via maximum likelihood (see under Frat). 

                      noise             
Setting       plies  (* 1000)   Frat    95% CI  
                                     ------------
supremo         2        0      2106 (2075, 2138)
expert          0        0      2035 (2011, 2061)
  -             0        7      1958 (1909, 2008)
advanced        0       15      1825 (1776, 1874)
  -             0       22      1643 (1594, 1691)
  -             0       31      1468 (1413, 1522)
interm.         0       40      1220 (1169, 1270)
cas.pl.         0       50       940 (888, 992)
begin.          0       60       714 (664, 764)


For the 0 ply settings, I fitted the following polynomial: 

 Frat = +2.0368E+03 - 8.93668*n - 4.33204E-01*n^2 + 3.560385E-03*n^3 

with n = noise * 1000.  E.g. this would imply that a noise of .029 equals about
1500 level.

Note that the bots played one pointers only. I do not know how noise for cube
decisions would affect its strength. 
Furthermore, I think its strength is 5-10 points higher than indicated due
to droppers/unfinished games.


greetings,

boomslang


                
___________________________________________________________ 
To help you stay safe and secure online, we've developed the all new Yahoo! 
Security Centre. http://uk.security.yahoo.com