Re: computer-go: Passing on a 9x9 board

[Matthew Corey Brown <bromoc@xxxxxxxxxxxxxxxxxx>]

On Tue, 12 Jun 2001, Florian Bantner wrote:

> Hi,
> I don't know enough from neuro-nets so this is just an idea:
> If the net thinks it should pass after 35 it's just silly and there comes
> darwin: mark it as died out species and let it die in peace for the
> next generation. A cut-off in the net only simulates the intelligence
> the net has to get by it's own and is not trivial in go. 

Yeah thats the rub, too little and it will think passing way early in the
game is perfectly okay. I realize passing maybe a good strat in some
cases. However i only want to not count the score if BOTH nets pass
early. Trying not to award due to luck. But also I don't want it to learn
that all passing is bad in the early game.

> But perhaps
> you could use as a cut-off the calculation-time from gnugo by itself
> and break calcluation off after - let's say - 10 seconds ?
> 
> regards
> Florian Bantner
> 
> 
> >I was wondering if anyone knows a good statistical number of moves where
> >both players passing results in an aborted game. Some times both of the
> >nurel nets playing against eachother pass after 20-35 moves, i have no
> >qualms about one player passing.. but if both do early on it has taken
> >upto 4 minutes for gnugo to try and calculate a score. Since i'll have to
> >rewrite the fitness function take in the new_score feature. I was thinking
> >of not even bothing scoring a game if they don't reach a certain number of
> >moves. So any idea what the threshold should be? 
> 
> ====================
> AXON-E interaktive medien
> mail: f.bantner@xxxxxxxxxxxxxxxxx
> fon: +49 / 941 / 59 98 54 0
> fax: +49 / 941 / 59 98 54 1
> web: http://www.axon-e.de
> 
> ----------------------------------------
> Der freie Veranstaltungskalender
> http://www.terminkalender.org
> 
> 

Matthew Corey Brown                             bromoc@xxxxxxxxxxxxxxxxx 
  "Death can not stop true love. All it can do is delay it for awhile."