Re: [gnugo-devel] endgame module for GNU Go

[Arend Bayer]

Paul wrote:

> We can use a planner if it is free software and can
> be compiled into GNU Go.  If we have this, then it is
> not difficult to provide an option to use an external
> planner (instead of built-in), for instance yours.

Note that it is very well possible to distribute your changes to
interface to an AI planer separately. Everybody who is interested could
download and install a patch (in addition to downloading an external
AI planer). If these changes turn out to make GNU Go stronger we can
discuss again how to integrate them into GNU Go.

We do the same with our own big experimental changes.

With regards to an endgame planer/breaking up the endgame into smaller
"subgames": Combinatorial Game Theory tries to do this from a different
angle. I think there are papers by Martin Mueller on how to use this
in computer go programs.

I don't know what an "AI planer" is, but maybe your approach could be
an inbetween between the CGT approach to endgame, (which needs a lot of
analysis of the subgames, with exponential growth with respect to the
"size" of the subgame) and a global brute-force minimax search.

I doubt breaking up the game in many small subgames (which I understand
is a requirement for your approach) will work in any other stage of the
game (but I would be pleased to be proven wrong).

Arend