Re: [gnugo-devel] Three stone games

[Morten Gulbrandsen]

Thank you for a polite reply.

How can we find a reliable measure for our computing efforts ?
what about this table:
Game nr     Black points White points   Black prisoners White prisoners         
        
        1       ???             ???             ???             ???             
        2       ???             ???             ???             ???             
        3       ???             ???             ???             ???             
This can be used to find out which colour advantage black has, and hence
the komi necessary                                              
        
In this table I expect 5.5 komi  for 50 % winning chance.  and a 19 x 19
board.
Other board sizes I expect gives other komi values. 


                                        
In chess we have numbers like                                           
        White wins      39.9%                           
        Black draws     28.9%                           
        Black wins      31.0%                           
                        99.80%  According to my information                     
                                                
                                                
                                                
In 9 x 9 Go I believe we could get something like  :                            
                
        Black wins      70.00%                          
        White wins      25.00%                          
        Tie             5.00%                           
                        100.00% 
without komi offset.                    
And this I'd like to investigate. When we know the numbers for 9 x 9,
then we 
can investigate if this is true for 19 x 19. 

                                                
                                                
Now we could add 1 Handicap stone                                               
        I suggest:                                      
        Black wins      75.00%                          
        White wins      22.00%                          
        Tie     3.00%                           
                100.00%                         
                                                
        The increased probability is the value of 1 Kyu  play strength          
                        
3 handicap stones                                               
        ???                                     
                                                
4 handicap stones                                               
        ???                                     
        
We know to beat 9 x 9 gnugo with 4 handicap stones, you must be 4
dan.                                    
I expect the difference of one increased Handicap stone to be linear. 

I'd like to do this, if it is helpful for you. gnugo versus gnugo, and
gnugo versus Morten Gulbrandsen. I only play as 13 kyu, so it could be
an even match.                                          

When I know how much komi offset for each board size, and each given
number of 
handicap stones is necessary for 50 % equal play, then I'm
satisfied.                                              
                                                
Or we could add some offset as komi. Then gnugo knows how many points it
needs  to just win                                              
In go we could vary much more than in chess, board size, initial komi
offset, handicap stones.        times.                                  
And gnugo could tell us something about the interrelationships between
these parameters.                                               
                                                
In chess we have some ELO figures to tell us how strong a player
is.                                             
In Go we have the kyu, shodan, dan, professional dan figures.                   
                        

usually 2100 ELO  = 1 dan.

I hope tuning and improving algorithms can tell us more about go. 
Tie is very seldom, achieving a seki is difficult, but interesting. 
Could gnugo be tuned to achieve larger sekis than common in human plays
?

I also find it very difficult to understand, why gnugo does not increase
its strength
with computing speed?  If gnugo plays against itself with different time
limits, 
then how many minutes will it need to play 1 kyu stronger ?

In chess we sometimes scale the difference of play strength with 
different time limits. the less experienced player gets more time.



I don't know what kind of statistics is helpful for the programmers, 
but I do hope we can use some statistics to find out how much the 
different gnugo versions have increased in kyu strength. Normally 
on a 19 x 19 board, the difference in kyu is 1 if the opponent needs 
2 handicap stones, for a 50 % winning chance.

Am I right ?

what is a kadoban game ?  

I've bin reading the gnugo task list and find that 
what I could do is 
        4.  Extend and tune the Joseki database.

I have some classical joseki book dictionaries, so if someone out there 
works on the Joseki database and if I can do this without being a dan
player, 
please tell me how to help. 

Yours Sincerely

Morten Gulbrandsen



Trevor Morris wrote:
> 
> These mean values of the score differences aren't really meaningful.
> GNU Go plays more conservatively when ahead.  That is, GNU Go is
> not tuned to maximize the point differential, rather to maximize the
> chances of winning.
> 
> The questions you ask make a lot of sense.  I don't have any hard
> numbers, but can share the following observations:
> 
> I've noticed some asymmetry when the same version of GNU Go plays
> itself even.  I've don't recall the details, however.
> 
> I did, back around 3.1.20 or so, play a series of 1-game kadoban games
> btw. GNU Go and itself.  It only rarely won 2-stone games, and never
> pushed itself to four stones (i.e. didn't beat itself at 3-stones).  The 
> number
> of games was pretty small, though.
> 
> -Trevor
> 
> >...
> >
> >        477     490     Sum of points
> >        18.35   20.42   Arithmetic Mean value
> >        18      17.5    Median  (gives white the advantage)
> >        12.14   15.11   Geometric mean value
> >
> >        14.47   14.68   Standard deviation
> >
> >I did some statistics on the material supplied.
> >hope this could be helpful for you.
> >
> >The achieved points are some kind of "territory"
> >White won 26 times. only the median gives white a
> >higher score.
> >
> >What happens when two equal versions of gnugo plays
> >9x9,  13 x 13 and 19 x 19 board sizes?
> >
> >Without handicap stones ?
> >Which komi is required for each size ?
> >
> >Yours Sincerely
> >
> >Morten Gulbrandsen
> >
> >
> >address@hidden wrote:
> >>
> >> Gunnar raised the question of how GNU Go would do
> >> in three stone games against 3.0. A few releases
> >> ago it lost about 65% of the games but now they
> >> seem to be evenly matched with this handicap. Here
> >> are the scores in a 50 game series.
> >>
> >>...
> >>
> >> Dan
> >>
> >> _______________________________________________
> >> gnugo-devel mailing list
> >> address@hidden
> >> http://mail.gnu.org/mailman/listinfo/gnugo-devel
> >
> >_______________________________________________
> >gnugo-devel mailing list
> >address@hidden
> >http://mail.gnu.org/mailman/listinfo/gnugo-devel
> 
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gnugo-devel