gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] strength against Many Faces


From: David G Doshay
Subject: Re: [gnugo-devel] strength against Many Faces
Date: Tue, 19 Apr 2005 11:13:20 -0700


On 19, Apr 2005, at 10:37 AM, David Fotland wrote:

Are the games played in a way that allows Many Faces to learn?

After a posting from you about a week ago, I have been meaning
to ask you this same question!

  It keeps the
learned positions in a file in the install folder called flearn19.dat. This file is read once when MFGO starts up, and is written after each completed
games with the new learned positions.

We do not restart MFGO between games. We set the internet play
to accept all matches and roll over from one game to the next as
soon as a game finishes.

MFGO will prefer to repeat openings where it wins games, and will try to
avoid openings that lead to losses.

This may explain why some of our clusters get different results than
others, although by now I might have guessed that they would have
converged after a few hundred games.

I intend to go back thru the game records and see if the winning
percentages are greater with one particular opening ... but it has
been hard to inspect them all. Maybe it is time for a perl script.


Cheers,
David






reply via email to

[Prev in Thread] Current Thread [Next in Thread]