Re: [gnugo-devel] The Module Formerly Known As Endgame

[Eric]

--- Gunnar Farnebäck <address@hidden> wrote:

> Warning, these comments are sometimes rather
> negative and unconstructive

Then what is the point in making them? Just to be
argumentative?

> Eric wrote:
> > What happens if we replace minimax search and
> > alpha-beta pruning by more recent search tree
> pruning
> > techniques, e.g. recent advances in AI Planning?
> 
> This is a key question. I have no idea but I would
> be very interested
> in the answer.

You will never know unless you find out.

> > Surely some Go games must exist in which taking
> prisoners leads to
> > winning. These are the games that the planner will
> look for.
> 
> Occasionally the players get into sharp positions
> with races to
> capture, where the player who loses his stones
> totally crumbles.

That is "occasionally", Gunnar. I am not trying to
specify all the details of playing Go to the search
program. That is the purpose of search is to do the
work.

> > Because we have a relaxed search goal now, we
> would be
> > able to solve 6x6 boards (i'm just pulling these
> size
> > increases out of thin air). Right? Because we're
> not
> > trying to win the game anymore. In the new
> scenario,
> > "solving" the 6x6 game board just means
> constructing a
> > plan for taking prisoners on that board. This
> > relatively simpler goal can (sometimes) be solved
> with
> > less effort.
> 
> That may be possible but it has very little to do
> with getting strong
> at go.

At this point I'm not trying to "get strong at go".
I'm merely trying to increase the size of the board
that my program can operate on.

> I'm sorry to be harsh, but from a go playing
> perspective this approach
> makes absolutely no sense at all.

You're not being harsh, just stating the obvious. It's
obvious from your comments that you are not following
my reasoning.

> GNU Go hasn't been exclusively pattern matching
> since version 1.2,

What else does it use?