[computer-go] thoughts on 100,000 cgos games

[Don Dailey]

I believe that if there is a way to learn from game collections, it has
to be centered around the mistakes.    We really get too focused on
"good moves" and try to make our program play "good moves" instead of
not playing "bad moves."    

In physics there is only degree's of HEAT - no such thing as cold except
as a way to express a lack of heat.

In games like Go, games are won because of mistakes - not because they
are produced by brilliant play.   When we say a move is brilliant we
don't really mean that.   What we really mean is that it is really easy
to play an incorrect move!

So in a way it's silly to try to learn from your won games.   You are
basically taking the games where the opponents made mistakes and
expecting to get information from YOUR play (when the loss was actually
caused by the opponents mistake, not by your follow up.)   Be we like to
pretend that the follow up move(s) caused the victory.

- Don


 

On Sat, 2006-05-06 at 13:08 -0700, David G Doshay wrote:
> On 1, May 2006, at 5:46 PM, Chris Fant wrote:
> 
> > Isn't #3 fixed by only learning based on games that SlugGo loses?
> > _______________________________________________
> > computer-go mailing list
> > computer-go at computer-go.org
> > http://www.computer-go.org/mailman/listinfo/computer-go/
> 
> I took my time replying to try and really think about this suggestion.
> 
> I can think of no solid way to benefit from an automated approach
> given only the information that I lost all of these games. I can think
> of ways to do this manually, going through the games one at a time
> and marking mistakes, but that is just too time consuming.
> 
> If you have better ideas I am very interested in hearing them.
> 
> Cheers,
> David
> 
> 
> 
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/