Re: [gnugo-devel] assignment with gnu-go

[vsteigvi]

It seems like a good idea. I just had a look to see if the Berlekamp and
Wolfe's, Mathematical Go was in University of Guelph/Waterloo/Laurier libraries,
but no such luck. I did get some papers off the citeseer search engine by Martin
Muller that look along similar lines. I'm going to have a read over those
tonight. Please feel free to make suggestions on details or other idea's

Thanks Vytas.


Quoting address@hidden:

> 
> > Sorry I think I accidentally sent my last message to your email and not
> the
> > "address@hidden". Funny, I can't find it in my sent mail.
> > 
> >I talked with my prof. and he seemed really happy if I could do something
> > that could benift GNU and saw no problems with copyrighting to FSF.
> 
> I have a specific suggestion which is to implement an endgame module
> based on the combinatorial games theory paradigm, a Go endgame is a
> sum of local games.
> 
> This point of view is advocated in the book of Berlekamp and Wolfe.
> Their specific algorithm is not practical because it is only valid
> for cold and tepid games, where each local game is worth at most
> one point. By the time a Go game reaches that stage, it is usually
> already decided. But their book is worth looking at for ideas if
> you can find it.
> 
> A first problem is to get a good algorithm for localizing the
> endgames.
> 
> Then I think you can use the metamachine to compute the values of
> local games. Not in the precise way that Berlekamp and Wolfe do
> but some approximation similar to what human players use, this
> move is worth 2 points in gote, 1 point in sente ...
> 
> Then once the local games are computed, to find the best move.
> 
> Let me know if you like this idea.
> 
> Dan
> 
> 
>