Re: [computer-go] KGS Tournaments: Format

[Don Dailey <drd@xxxxxxx>]

First, I'll talk about cheating protocols.   Of course I am quite
jaded coming from the computer chess world where cheating and paranoia
over cheating have been problems in the past.

It seems ludicrous to me that anyone would cheat, especially over 
such a low profile tournament with no money involved,  but it 
seems ego is more motivation than money.   It will inevitably
happen I'm afraid. 

I don't see a very good solution.  The TP protocol is the best idea,
but it has many problems.  If this solution is chosen, I can easily
send a statically compiled version to a TP, but this might not be
possible for some programs like Sluggo.  Also, my program benefits
from running on a fast computer, so I will squabble about TP's with
slow computers!  See what a can of worms this can be?  

If you want to EXPERIMENT with TP tournaments, I will participate by
sending someone a static binary of my program and we can have some
fun.  I suggest you try to support the big 3 platforms however,
Windows, Mac and Linux.

The reproducible program idea is not very good.  There is no way a
multiprocessor program or a networked program could compete.  You
should not require a program to be reproducible in order to compete.

> A real, cheat-proof system for online tournaments has not been found yet AFAIK.

How very true!

A good try is a uniform platform tournament where the hardware is
selected in advance and binary's are distributed to trusted parties.
But this is quite restrictive unfortunately.


- Don






> FREQUENCY OF TOURNAMENTS

I like once per week or as often as reasonably possible.  To make this
easy to manage, the tournaments can be fully automated including
scoring.  With chinese, it is easy to score and programs must obey the
kgs-genmove_cleanup protocol or take their own chances.  At the end of
the game, when both computers have passed (due to kgs-genmove_cleanup)
the server determines the score which stands.  This score cannot be in
error as long as one of the programs properly obey the protocol.

> DAY OF THE WEEK

I'm not fussy.  Try to accomodate as many as possible.  Weekends seem
to be good times for most people.

> TIME OF DAY

Same as above, even more so.  Mix up times to accomodate everyone.

> BOARD SIZE

My program can play on any size board, but is optimal at 9x9.  
But 19x19 is the "real" game.    

If the tournament can be fully automated,  why not have both a 9x9 and 19x19
tournament at the same time?    With liberal time controls even slow
programs could play on both at the same time.


> TIME LIMITS

Have automated tournaments every week,  each week could be a different
time control.  Have a mix of fast and slow.  


> PAIRING SYSTEM

If it becomes possible,  round robins are best.   Double round robins
even better.   Since the server can accomodate hundreds of simultaneous
games,  it is not impractical to have farily large round robin tournaments.

Otherwise, swiss is far more practical than elimination.






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