[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] future KGS Computer Go Tournaments - two sections?
In message <BAY9-DAV16C2F18C63F2E517D67B6A9C100@xxxxxxxxxxxxxxxxx>, Vlad
Dumitrescu <vlad_dumitrescu@xxxxxxxxxxxxxxxxx> writes
Hi all,
I hope I may throw in my 2 cents on the issue. Probably nothing new, but
hopefully in a different light.
I broadly agree with Vlad - but it seems I differ from him on several
relatively unimportant points ...
I think that there is one question one would like to answer before diving
deeper: what is the purpose of computer-game tournaments?
If the answer is "to crown the best playing program", my feeling is that any
program should be allowed to enter, in any amount. I assume an
"all-play-all" tournament, so that no competitor is pushed down in a lesser
league just because there are many entrants.
If the object is "to crown the best playing program", and we allow any
programs to enter in any amount, the leading contenders will enter
dozens of copies. This will make the numbers too high for Round Robin,
so we will be using Swiss. And the results table will be absurd, with
some programs doing nothing but play copies of themselves.
Also, this will tend to reward programmers who can borrow a commercial
network for the weekend, to run all their copies on.
If the answer is "to crown the programmer/team that produced the best
playing program", then it is of course unfair to allow a lucky newbie to
stumble onto a combination of tuning parameters and get credited without any
nights lost and sweat and tears.
I have nothing against rewarding lucky newbies.
If someone applies some new technology to write a bot, and thereby
achieves something which can beat MFoG, while putting in less than 1% of
the effort that David has invested, that'll be great!
It is in the latter case that we have the main grey zone and the most
difficult choices. And in my opinion, the two leagues proposed by Nick are
nicely reflecting these two ways to look at things. Maybe the "best program"
league entrants could run anonymously, only TD and respective authors
knowing who is behind a handle?
I think I am going to go with the two leagues.
best regards,
Vlad
p.s. There is a grey zone in the former case too: what if a rich person buys
a supercomputer and runs GnuGo out of the box on it? Maybe right now the
programs aren't very scalable, but in the near future they might. Then
hardware might matter more than algorithms... But I feel this one is much
less of a problem.
I don't think it is a big problem, and I am not planning to do anything
about it.
It's not just that the programs aren't scalable, it's that the whole
task isn't scalable. Give a program more time and it doesn't really
know how to use it. This is what makes SlugGo interesting. The SlugGo
team are doing their best to see how much improvement they can get by
applying extra computing power. I hope they will forgive me if I say
that despite their efforts, the increase in strength is not
proportionate to the computing power applied.
Nick
--
Nick Wedd nick@xxxxxxxxxxxxxxxxx
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/