On Sat, Apr 03, 2010 at 11:26:55AM +0800, Scott Christensen wrote:
> Thanks for all the info. After a bit of thought capturing rules are
> very simple really, I just test every group adjacent to a new move.
> Using the CGOS tromp/taylor system I don't need to worry about
> life/death at the end-game. For internal state the only other
> essential would be keeping track of the 'ko rule'. Most typical
> API's have a lot of redundant check state functions which aren't
> included in GTP so a bit more testing up front is needed on my part.
>
> Thanks Petr for your excellent summary of go engine techniques:
> http://pasky.or.cz/~pasky/go/compgo-r2.pdf
You are welcome. :-)
Please note that the presentation covers only the successful
undertakings - it is heavily focused on the Monte Carlo approach
(currently by far the most promising) and almost entirely omits other
approaches tried in the past, such as alpha-beta/lambda/PN search
+ a lot of knowledge or Neural Network based players.
> I'm working on a 'mothering' technique where abstract understanding
> and common sense approaches are built up from the 30kyu level in order
> to achieve very tight tree pruning,etc... its strange that we would
> not expect a human beginner to be able to learn anything from a 9-dan
> game, but that's what we often expect a computer program to train
> with.
I recommend to look in more detail into GNUGo inner workings for
similarities since that's pretty much the approach taken by the
"classic" programs.
--
Petr "Pasky" Baudis
http://pasky.or.cz/ | "Ars longa, vita brevis." -- Hippocrates
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
