|
Joined on 03 мар 2004
Total posts: 5
|
Couple ranking system
18 марта 2004 г. 6:46
I'm probably repeating previous suggestions but www.scrutelle.com has a package for collecting marks that scrutineers (anyone ?)could use.
As for limiting the degree to which acouple's marks can change as a result of a single event's performance - sounds like some kind of "S" curve function is required (we know - the shape that describes a population's growth depending on a limited but renewable food source (!)).
|
|
Joined on 12 авг 2003
Total posts: 84
|
Couple ranking system
18 марта 2004 г. 10:55
The amount by which your score can increase from a single 1/1 comparison is already limited to the value usually described as "k". So Bryan being in the first round will have increased a few people's scores by 16 points, but in the grand scheme of things these errors probably won't matter.
Much more of a problem is that Bryan's score could have been affected very drastically because the system thinks he lost against 100s of people, each subtracting 16 points from him.
Maybe it is possible to describe a cut-off if someone is beaten by several people with 500 points less than themselves they are withdrawn from the competition. I'm not sure this is practical... An easier thing would be to limit the total number of points lost in a competition to a certain value, say 50 or 100...
Robin
|
|
Joined on 12 авг 2003
Total posts: 84
|
Couple ranking system
18 марта 2004 г. 11:10
I had a look at the new feature, and it seems the system over-reacts due to the large number of couples in one comp. Here's an example from the UK open:
94 [23] Marco Bevilacqua & Alessandra Ippolito Italy 2829 2274
94 [23] Francesco De Candia & Sabrina Laconi Italy 2525 2306
94 [23] Aerjen Mooijweer & Monique Broekmeulen Netherlands 2617 2299
94 [23] Juris Uzegovs & Jelena Rakamina Latvia 2467 2309
All 4 couples finished at the same position. In the end the one who previously had the highest rating now has the lowest of the four. This is due to the fact that he deviated most from his expected score and thus was corrected most. In chess people rarely play more than 7 or 8 games in a competition, so this does not occur.
My suggestion: Reduce the k-factor in general or scale it down with the size of the competition...
Robin
|
|
Joined on 14 фев 2001
Total posts: 1 976
|
Couple ranking system
18 марта 2004 г. 22:57
I agree, that we should scale winnings in big comps, as winning with 200 weak couples gives more points than winning with one good.
Question is by what 1/cnt or 1/sqrt(cnt) or 1/log(cnt) ?
Any idea?
It would also be quite interesting to ignore couples who lost too many points (as it was probably due to the mistake), unfortunately that would work if starting assesment was close to right.
It's interesting that in general system behaves quite right for top ones and is a bit unstable for weaker couples
|
|
Joined on 12 авг 2003
Total posts: 84
|
Couple ranking system
24 марта 2004 г. 14:43
I would expect the system to be more random for couples lower down, mainly because their performance is very variable. Also especially in the large competitions where you are not given all the marks, the difference between 1st and 2nd round could be 50 places, but only due to 1 mark.
The top dancers usually are competing against the same people with similar results so you would expect the system to work better.
Anyway, formula for scaling large comps... I suggest you try 1/ln(cnt), where cnt is the number of couples within the cut-off you use (not the total number of competitors). I think that should stop all over-shooting in large comps, though it may be too harsh.
Robin
|
|
Joined on 06 июн 2004
Total posts: 2
|
Couple ranking system
6 июня 2004 г. 12:16
Not sure why this thread has died, but I'm glad I found it, and would like to say how great I think the Ranking what-you-may-call-it thingy is!!
The only dancing site on my favorites, and a greatly usefull page.
Thanks
|
|
Joined on 14 фев 2001
Total posts: 1 976
|
Couple ranking system
7 июля 2004 г. 15:42
It's not dead :) We were just very very busy during the last 2-3 months, but hopefully we will restart it this summer
|