In message
<[email protected]>, Erik van
der Werf <[email protected]> writes
On Tue, Sep 7, 2010 at 12:45 PM, Nick Wedd <[email protected]> wrote:
Somehow I feel that these kgs tournaments are for an important part
decided by the kgs paring algorithm. It's annoying; the olympiad has
become a hardware contest and the kgs tournaments a lottery. Hardly
any good tournaments left. (Ok, maybe I'm exaggerating a bit, it's
still fun to watch :-))
I will take this as praise for the way Jukka ran the EGF Computer Go events
in Finland last month :-) They were round robins, so no complaints about
the pairing, and participants had to use the hardware provided, or carry
their own into the room, so no connections to supercomputers.
Yes, that was the exception I had in mind :-)
As for the KGS events, I have considered alternatives.
I could do the draw myself. It's hard work; it takes time (slowing the
tournament down); it requires all the operators to edit their config files
before the start of each round (slowing the tournament down more); and I
would make mistakes. Ok, the KGS algorithm makes mistakes, or at least
questionable decisions, too; but no-one can accuse it of bias.
I could use a round robin format; or double, or triple, round robin, as
supported by KGS. The problem with a round robin is that the length of the
event is then determined by the number of entrants, and I don't want a
tournament to take more than eight, or less than around three, hours. So if
it's a round robin, I would need to be able to determine the time limits
and/or the multiplicity of the round robin _after_ all the entries were in.
But some people like to enter at the last minute, and others enter in
advance and then don't show up. And if I specify a cut-off time for
entries, and then remove all those who have entered and not shown up, and
only then "create" the tournament, it's going to take me at least ten
minutes to check that I have created it with all the right settings and
entered all the right programs into it. Meanwhile all the participants are
kept waiting.
Any suggestions?
For 9x9 I'd prefer double round robin, to get rid of the color bias.
For large boards single round robin is fine.
As things are, I don't think double round robin will get rid of the
colour bias. The 19x19 Computer Go event in the EGC was double round
robin, you can see the results at
http://www.gokgs.com/tournGames.jsp?id=526&round=3 , and in both rounds
3 and 4, Many Faces played pachi, Many Faces taking white both times.
I reported this anomaly to wms, and he agreed that this could easily be
fixed (with no need to change any user interface and no risk of it
interacting with anything else). He then told me he had fixed it, but
it would not come into effect until the next rebuild of that part of the
server. It appears not to have come into effect yet: on Sunday, PueGo
played Zen9 in round 6 and again in round 20, PueGo was black both
times.
I don't mind last minute changes to the time control because kgs-gtp
takes care of that (but maybe that's different for other programs?).
Does kgs support McMahon? It would be nice to reduce the number of
unbalanced games.
KGS supports McMahon. But this only makes sense for 19x19 games (KGS
ratings are earned only from 19x19 games); and even then, many 19x19
bots do not have ratings, and cannot easily acquire them.
I think even Swiss could be ok, if only the pairing was somehow a bit
smarter...
I do not expect to see "smarter". I still hope to see non-random
colours for repeat pairings.
Nick
--
Nick Wedd [email protected]
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
