Re: computer-go: Results on self play vs recorded games

[Markus Enzenberger <compgo@xxxxxxxxxxxxxxxxxxxxx>]

On Thursday 21 August 2003 21:11, Peter Drake wrote:
> information.  Is there a consensus on how best to get
> some variety in testing?  Doing several training runs

add a small amount of noise to the evaluation, just enough 
to get different games.
Alternatively, you could use an opening book to choose the 
first couple of moves or test against a program that has 
some variety in its opening play.
I had good experiences with using GNU Go 3.0 on 9x9 (later 
versions have less variety in the opening).
There are tools for automatically playing a series of games
with programs supporting GTP.
e.g. twogtp.jar, which is part of GoGui and also detects 
duplicate games, however the newest version is not released 
yet, you have to get it via CVS.

- Markus