[computer-go] Future KGS computer go tournaments

[Nick Wedd <nick@xxxxxxxxxxxxx>]

I would like to thank all those who responded in the various threads 
about KGS Computer Go tournaments.  In this email I try to summarise 
your views and say what my plans are.

PREVENTION OF CHEATING

People agree that prevention of cheating, by those determined to cheat, 
is difficult or impossible.

The "Trusted Person" idea attracted little interest.

The idea of requiring programs to play reproducibly attracted more 
interest and more opposition.  Those whose programs can be persuaded to 
play reproducibly were more inclined to favour it.
    Personally, I tend to be opposed to requiring programs to play 
reproducibly.  I would not want to exclude programs that use particular 
techniques.  RNGs can be seeded, and time constraints can be recorded 
and reproduced;  the big problem is concurrency.  I think that we will 
be seeing more of concurrent programs in the future, and it would be a 
pity to bar them from a tournament.

My plan is to have frequent, probably monthly, computer Go tournaments 
on KGS, with no significant rewards for winning, no attempt to prevent 
cheating, and I hope no incentive for anyone to cheat.

I think that we will also see a sponsored computer Go tournament on KGS 
before the end of this year, with a prize for the winner.  It will be up 
to the sponsor to choose what methods should be used to deter cheating. 
I will do my best to cooperate with whatever methods a sponsor chooses.


UNIQUENESS OF ENTRANTS

I thank Gunnar Farnebäck for his suggestion, which I shall accept. 
Gunnar wrote:

"My suggestion is to accept a derivative work if
"
"1. All copyright holders agree that they want it to enter the
"   tournament and
"2. the copyright holders can convince the tournament director
"   that it is indeed unique enough to be interesting.

The snag is that this imposes a burden of authority on the Tournament 
Director.  But I think this is inevitable.

For tournaments for which I am TD, I expect to decide as follows:
    If two entrants both use the same fuseki library, and the 
copyright-holder of that library (if any) wants both to enter the 
tournament, I shall not see this as a reason to bar either.
    If two entrants both use Thomas Wolf's GoTools, and Thomas wants 
both to enter the tournament, I shall probably not see this as a reason 
to bar either.
    If two entrants are both based on GNU Go, and I am not convinced 
that they are greatly different, I shall ask a representative of GNU Go 
to choose just one of them.

(Does GNU Go have an official spokesman?  If I need a decision carrying 
the authority of the GNU Go development team, whom should I ask?)


FORMAT OF TOURNAMENTS ...

FREQUENCY OF TOURNAMENTS

Monthly was favoured.  I shall go with this.  I hope to announce the 
next one this week.

DAY OF THE WEEK

The only person to respond on this preferred weekends.  I shall go with 
this, which will force me to choose weekends where I do not have other 
commitments.

TIME OF DAY

No-one from Asia responded.  Nevertheless, to be fair I shall try to 
rotate time to suit all three "major timezones".  This will be easier 
when we have more people able to act as judges if a result is contested.

[ A "judge" is a person with the power to assign a result to a game. 
This power will only normally be used when, at the end of a game, the 
two bots fail to agree on which stones are dead.  I believe that at 
present, only three people are able to act as judges on KGS - two in 
western Europe and one in the western US.]

Another way to be fair to all timezones is to have a tournament which is 
continuous around the clock, so as to be equally inconvenient for 
everyone.  This is inconvenient for human observers, but not for 
programs, which can be left running when their owners are asleep.  (When 
you do this, put "reconnect=t" in the KGSGtp command line or 
configuration file.)

BOARD SIZE

9x9 was most popular, then 19x19.  Only one person asked for 13x13.

I plan to alternate 19x19 and 9x9.

TIME LIMITS

One request for constancy, one for variability, one for Canadian.

For each board size I plan eventually to use not more than two, and 
maybe only one, time system.  But while we are finding out what works 
best, I may try a variety of time systems.

As the person responsible for the schedule, I prefer Absolute time. With 
any other system, you can't predict when a round will end, so you can't 
know when to start the next one.  This is not such a big deal with 
over-the-board tournaments, if one game uses lots of overtime you just 
keep everyone waiting until they have finished and then draw the next 
round.  But KGS's automated tournament system requires to be told in 
advance when each round will begin.

For tournaments that take place in a single session, I shall stick with 
Absolute time.  For tournaments played at one round a day, or "around 
the clock" with generous gaps between rounds, I shall try Canadian 
overtime.  We may need more judges before we run around-the-clock 
tournaments on KGS.

PAIRING SYSTEM

Several people requested Round Robin (all-play-all) and Double Round 
Robin.  The KGS system does not yet support these.  I believe that 
William Shubert plans to add the Round Robin format.  He may also 
consider Double Round Robin, with its constraint that the second time 
the same two play, it is with colours reversed.

RULES

I forgot to mention this earlier, and no-one else has.  I am sticking 
with Chinese (area) rules.  They make the endgame easier for everyone, 
particularly for the judge.

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