[Computer-go] Semeais

[Thomas Wolf]

On Fri, 14 Jan 2011, Stefan Kaitschick wrote:

> ...
> When you look at bot behaviour in the face of multiple semeais the reaction 
> is "there has to be a better way".
> But trying to isolate semeais from the rest of the board gets you back into 
> the mud that go programming was stuck in before mc.

A new approach which addresses exactly this and related problems is described
in http://lie.math.brocku.ca/twolf/papers/IEEE10.pdf .  It sets relative
strengths of chains and influences at empty points in relation to each other
and computes them iteratively in terms of each other until iterative changes
are small enough or a limit of iterations is reached.  The described method is
static, i.e. does not contribute to combinatorial explosion of moves and
sequences to be investigated. It comes with a static computational cost but it
does not seem to be higher than the cost involved in heavy playouts. For
semeai positions the described static evaluation gives good predictions of
which chains are stronger and what are the best moves to be done
first. Because the method is static, it less useful in areas where chains have to be
sacrificed or where the shape and size of eyes is critical or when ladders
matter, so in short, when moves are needed to be made.

> You need to be able to formulate improvements within the mc paradigm.

The mentioned method needs to be merged with a mcts program for further
testing and development.

> Something along the line of finding move equivalency classes by statistical 
> means.
>
> Stefan

Thomas

>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>