Re: [computer-go] future KGS Computer Go Tournaments - two sections?

["Jim O'Flaherty, Jr." <jim_oflaherty_jr@xxxxxxxxx>]

Don,

I am not persuaded to your point of view...at least yet.

What about the case where a person takes an existing engine authored by someone else where the
engine primarily drives from strategy configuration files (very little code based in Go logic). 
So the submitter is not an engine author, he is a configuration file author.  Does this
configuration author get to submit an entry?

After thinking about that, consider all the different possible cases between the two poles,
completely code based engine versus completely configuration driven engine.  Exactly how do you go
 about distinguishing what is sufficiently unique about the "incremental" author's contribution
over and above the "originating" author's contribution such that it legitimizes an entry?  Add to
this, any program that learns as it plays (pointed out elsewhere).  If I were to obtain a copy of
that program, spend enormous time training it, why would it not be appropriate to submit an entry
based on my training effort?

Honestly, I don't think you (or the tournament organizer) want to worry about this level of
distinction.  If you simply make the tournament have a nominal fee and restrict a single computer
based entry per person (not author), most everything will work.  And if you decide to submit more
than one entry, you are going to have to get some buddies of yours to represent your other
entries.  There is nothing saying the person submitting an entry must be the
originator/configurator/brains behind the submission.  This enables all sorts of "entry
configurations" across a spectrum of implementation styles.  It's not perfect.  That is not
obtainable.  It's good enough.

I think the fixation on fair is not serving the overall purpose, to forward computer Go in a fun
and stimulating way.  There is no fair.  There is just games between submitted computer Go
programs with some level of statistical significance.  ALL THE REST IS JUST HUMAN DRAMA!

If you model the tournament the way I am suggesting, when someone comes in with a cluster with
which to compete, you can be sure the next tournament will have more clusters.  THIS IS A GOOD
THING, not bad.  Attempting to "level the playing field" perverts the search space.  If you are a
genius and you figure out how to make a killer program with much less processing power required,
you will clean up with a smaller cluster/hardware investment.  The more you rely upon brute force,
the less likely you are to win long-term when a more effective strategy is found and exploited.

Besides, most of this is moot until the submitted programs start playing much more consistently
and in the very low kyu (less than 5) range.  Until then, anything less than statistical analysis
appears to be mostly luck (to high a degree of randomness).


Jim


--- Don Dailey <drd@xxxxxxxxxxxxxxxxx> wrote:
> 
> Similarity is not the issue, authorship is.  Because as I have already
> pointed out I have several very disimiliar programs but I doubt I
> would be allowed to get away with entering all of them in the same
> tournament.
> 
> - Don
> 
> 
>    From: Christoph Birk <birk@xxxxxxxxxxxxxxxxx>
>    Date: Wed, 11 May 2005 12:59:15 -0700 (PDT)
> 
>    I am aware that this does not help with the current discussion but,
>    after all this pointing out the randomness in computer-Go tournaments
>    I would like to say that the same exists for human tournaments.
>    When you have ten 8-10 kyu players in a round-robin tournament,
>    how often do you think the (single) 8-kyu player will win the 
>    tournament ... that's just how life is.
> 
>    Christoph
> 
>    ps. like Marco I am NOT suggesting to allow indentical copies 
>        of the same program.
>    _______________________________________________
>    computer-go mailing list
>    computer-go@xxxxxxxxxxxxxxxxx
>    http://www.computer-go.org/mailman/listinfo/computer-go/
> 
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
> 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/