Re: computer-go: Programs learning to play Go

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

At 09:47 AM 8/21/2001 -0700, Matthew Corey Brown wrote:

> The one thing in my opinion that makes it different from backgamonn and
> checkers (blondie24) model, is passing.. THe other games there are a set
> condition that determines the end of the game. Not so in Go, where the
> players them selves choose when the game ends. (Well there is the rare
> condition when there are no legal moves left) The trick is to seprate the
> pasiing desicion from the best move descion. And this is what i am
> planning once i finish the rewrite of my NN computation program.
No.  Treat a pass just like any other move, and pass when the pass is
the best move.  This is what Many Faces does.  There is no need to separate
the passing decision.


> Put other intracatble problems (like TSP) are manageable with social
> learning techniques. I do expect to take many 1000's of generations to
> be even slightly competent as the dimensionality of the net is so
> high. But social learning speeds up the training of nueral nets. (I expect
> the 1000's of generations with social learning)
I'm don't think that a net with inputs of stone positions can ever learn to 
play
go with any understanding of simple tactics.  I'll make the following 
challenge:

After you train it up until you thing it is slightly competent, play it 
against the
Larson algorithm.  I expect it will lose and so demonstrate its incompetence.

The Larson algorithm is:  Pick the string with the fewest number of liberties
(choose randomly if there are ties).  Fill a liberty of that string, chosen 
at random.
If this move is illegal, pick another one.  If you program doesn't 
understand liberties and
eyes, it will lose to this trival algorithm.

If you want a slightly tough opponent, choose the liberty that has the 
highest number of adjacent
empty points rather than a random one.  But I think the random algorithm will
beat the neural net.

This simple test is what convinced me that my first go program, based on 
influence,  was
the wrong approach, and that tactics is much more important than shapes or 
influence.



> Matthew Corey Brown                             bromoc@xxxxxxxxxxxxxxxxx
>                   Happiness is a dry place to live.
David Fotland