Re: [computer-go] Just for fun

[Christian Nilsson <christian.nilsson@xxxxxxxxx>]

On Mon, 14 Mar 2005 18:58:52 -0000 (UTC), john@xxxxxxxxxxxxxxxxx
<john@xxxxxxxxxxxxxxxxx> wrote:
> I'd suggest only including those games where the human plays it out fully.

Most of these "anomalies" will have less effect if enough games are
played. Unless they appear way to frequently of course...

Anyway, the program I wrote wasn't a serious attempt at the ranking
problem, merely a small test. I believe, based on this test, that one
can estimate a 9x9 rating _somewhat_ correctly using only the game
results of one player...if given enough games against people with a
stable rating. :)

Any serious attempt would include checking for undo and making sure
the scoring was properly handled. The early resign issue would be
difficult to handle since some people actually do resign when losing
against bots..although this might be rare. One would also like to
support handicap games. This is however a more difficult problem as
the value of X handicap stones is somewhat uncertain.

/Christian
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/