I do not see the point on having equal hardware tournaments requiring
clusters to get a fair competition.
I am not interested in solving engineering problems in computer
science. I am interested in programs that play beautiful go with as
little computation as possible.
We can run a tournament on old single core 500 Mhz Prozessors using
really long thinking times and get the same quality of games that a
cluster would do using normal time limits.
Why exclude hobby programmers from fair competitions. Go programming
is thanks to MCTS now feasible as project for young programmers who is
learning by doing.
The truth is it is impossible to have one single type of fair
competition that makes everyone happy. I would suggest two different
classes of competitions:
A) Laptops
B) Unlimited
where B might be unnecessary. Also I do not think machines have to be
identical. The randomness of game outcomes is larger than the benefit
of having a slightly faster machine in a tournament with few games
played.
When it goes for B we already have CGOS as the ultimately fair and
free option. Anyone can connect to it and play as many games they want
and with a large number of games luck is also eliminated.
It is very strange that tournaments with 10-20 games played is seen as
a test of which program is the strongest.
I know David often test his monster machines on CGOS but I rarely see
any other big cluster playing on CGOS. In my opinion if it does not
play there it really does not exist.
But maybe we need to coordinate testing for the different sizes for
CGOS. Like having one week of 9x9 , then one week of 13x13 and so on.
It is currently very hard to get consistent rating on CGOS (9x9) for
strong programs.
Anyway if someone whould like to prove that their program (or cluster)
is currently stronger than Valkyria. I am running the version
Valkyria3.5.14-4cx on CGOS which is basically running 4 threads on 2.8
Ghz. It uses the opening book I developed for Kanazawa, which is very
different from older versions that played on CGOS. I am not talking
about getting a higher rating, but simply win more than 50% of the
games over a longer period of time.
Magnus
Quoting David Fotland <[email protected]>:
I guess it depends on what your goals are. If you want to compare
programming skill on equal hardware, why exclude the people who have shown
the additional skill to make a cluster work? You might as well restrict to
single core machines, since some programs don't support SMP either.
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
