In message <[email protected]>,
Jonathan Chetwynd <[email protected]> writes
had you considered weighting for time?
ie tournaments with longer time intervals might get higher weighting,
cf some human dan ranking ladders which do not include fast play.
That is a good idea, though it will work in the opposite way from giving
more points for tournaments with more rounds.
it has been brought to my attention that whilst humans like relatively
long time periods, and require them to play well, there is currently a
bias in computer tournaments towards fast play.
I don't know about a "bias towards fast play". Bias relative to what?
To what humans prefer?
In organising bot tournaments, on KGS and elsewhere, I used to come
under pressure to use slower time limits. This seemed odd, as the
programs didn't know what to do with the extra thinking time, you could
give them twice as much time and the moves they found were hardly any
better. Their algorithms did not scale.
But in the last four years, since MC programs started to appear, things
have changed. All good bots now use MC methods, which do scale. The
more time, and the more processors, a program has, the better it plays.
But now, I come under pressure (from programmers) to use faster time
limits.
This seems odd to me. Maybe the continuing advances in processor speed
are relevant.
Nick
On 2 Jul 2010, at 10:37, Nick Wedd wrote:
This is to announce the start of the annual "KGS bot championship".
From now on, in each monthly KGS bot championship, points will be
awarded for the first few places - there might be four points for
first place, two for second, and one for third. These points will be
summed over the year, and the program with the most points will be
declared "annual champion" at the end of the year.
I am still working on how to award the points. The number awarded in
each tournament will depend mostly on the number of players, also to
some extent on umber of rounds, and board size.
As we are already in July (with the next bot tournament this Sunday),
2010 will be an experimental half-year, with only the results from
July-December counted.
Nick
-- Nick Wedd [email protected]
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
--
Nick Wedd [email protected]
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
