RE: computer-go: Perl Module for next move.

["Grajdeanu, Adrian" <adrian.grajdeanu@xxxxxxx>]

This is not going to work too well. For capturing dead groups means loosing
points. A program doing this will not be quite intelligent I would say.

> -----Original Message-----
> From: Fant, Chris [mailto:chris.fant@xxxxxxxxxxxxxxxxx]
> Sent: Monday, 04 June, 2001 12:40
> To: computer-go@xxxxxxxxxxxxxxxxx
> Subject: RE: computer-go: Perl Module for next move.
> 
> 
> What's so hard about that?  If you don't want to do any 
> reading to determine
> the status of the remaining strings, can't you just use GOE 
> rules?  Consider
> all strings to be alive at the end of the game.  The network 
> will have to
> learn to remove dead strings from the board to before passing 
> if it expects
> to get credit for killing those stones.
> 
> 
> -----Original Message-----
> From: Grajdeanu, Adrian [mailto:adrian.grajdeanu@xxxxxxxxxxxxxxxxx]
> Sent: Monday, June 04, 2001 11:28 AM
> To: 'computer-go@xxxxxxxxxxxxxxxxx'
> Subject: RE: computer-go: Perl Module for next move.
> 
> 
> Very similar approach I had as well. Where I got somewhat 
> stuck is when I
> play two networks against each other and try to figure out 
> the Go score. Did
> you tackle this phase yet?
> 
> > -----Original Message-----
> > From: Matthew Corey Brown [mailto:bromoc@xxxxxxxxxxxxxxxxx]
> > Sent: Friday, 01 June, 2001 18:09
> > To: 'computer-go@xxxxxxxxxxxxxxxxx'
> > Subject: RE: computer-go: Perl Module for next move.
> > 
> > 
> > A 3 layer(high node) neural network, The weights start off 
> random, the
> > inputs are the board locations, 0 for empty, 1 for your color 
> > and -1 for
> > the opponent. The output is one value the current score (Not go
> > score, the score the net gives the board)  you take the 
> orgional board
> > position then go through all possible moves to find a higher 
> > score. you
> > take the highest score move and use that on your turn. No 
> > higher scoor you
> > either pass or not. You start with some number of nets with 
> randomized
> > values and have them all play eachother.. you total everyones 
> > score (this
> > time the Go score) then you take the top 50% scorers from 
> that.. cross
> > breed them using genteic algorythms, and then apply mutations 
> > to each one.
> >  repeat the prosses for a long time. And after 1000 
> > generations or so the
> > neural net may learn the game well enough.
> > 
> > would be interesting to see at which generation strategies 
> > begin to form.
> > 
> > theres a checker player program that used around 800 nodes 
> > and evolved the
> > same way thats real good from what i understand after about 
> 7 weeks of
> > evolving on a pIII 400
> > 
> > 
> > On Fri, 1 Jun 2001, Grajdeanu, Adrian wrote:
> > 
> > > Amaizingly, I want to do the same thing...
> > > To analyze the table you use parts of the gnugo? If not, 
> > how do you solve
> > > it?
> > > 
> > > Adrian
> > > 
> > > 
> > > > I'm interested in giving a function the current board, then 
> > > > getting back
> > > > an array of answers either the resulting borad after dead 
> > peices are
> > > > removed or an illeagle move. Perl is needed cause I have 
> > a mishmash of
> > > > computer architechures to use Genetic algorythms to 
> > develop a neural
> > > > network to determine the best move at that moment. I want 
> > to watch the
> > > > computer develop its own stratagems.
> > > > 
> > 
> > Matthew Corey Brown                             
> > bromoc@xxxxxxxxxxxxxxxxx 
> >   "Death can not stop true love. All it can do is delay it 
> > for awhile."
> > 
>