|
Joined on 03 Mar 2004
Total posts: 5
|
Couple ranking system
Monday, 15 March 2004 17:03
What you are saying Dave harks back to what I was suggesting earlier.
Without getting too mathematical there is an ad-hoc factor in the ELO system which could be turned into a variable to reflect event strengths. As you say, this variable should in turn be calculated from the strength of the competitors in the event. One just (!) has to decide on the formula for it and have a good guess at the initial event strengths. (Mathematical suggestions to be continued via e-mail ?)
|
|
Joined on 14 Feb 2001
Total posts: 1,972
|
Couple ranking system
Monday, 15 March 2004 17:35
Awaiting ideas :) Have some other work to do, but I will fit dancesportinfo when I know what to do :)
Email is fine.
|
|
Joined on 12 Nov 2003
Total posts: 41
|
Couple ranking system
Monday, 15 March 2004 20:45
We appreciate that you must have put a lot of time into trying to work this out. Some competitors came up to chat yesterday at a comp and said they had been following this saga with great interest and were very supportive of our ideas and keen to see something come of them. Unfortunately they were having difficulty registering so that they could participate in the forum! From what I gather, there is definite support for having results at pre-champ level and below reported. Opinions are divided about the possibility of a ranking system though. As a competitor I am very anxious not to upset anyone!!! However if we can get the results up on the site the powers that be may realise that we are not trying to be subversive - just encouraging!
|
|
Joined on 12 Aug 2003
Total posts: 84
|
Couple ranking system
Tuesday, 16 March 2004 09:50
I can't see what anyone has to lose by there being an "unofficial ranking" on dancesportinfo. It's not like the rankings are meant to count for anything, it's just a bit of fun, and i think a lot of people will enjoy seeing their score rise over the years.
It would be good to hear from some of the "lurkers" that are apparently reading the board...
OK, my three further suggestions to try to improve the ELO style ranking:
- calculate intial scores for new dancers from the average of the round they went out in in their first event. Also have a variable in the database counting number of events/competitor and if that is lower than 3, increase the variability of the score (ie increase the K)
- Have a cut-off to the probability distribution. E.g. if the difference in rating is greater than x (say 400), make the probability 1, rather than 0.9. This will avoid the effect of large comps inflating scores as the eventual winners won't get a couple of points for every single beginner that entered them.
- If there is an event with no participants with a prior ELO score, you might have to flag that up and give people an initial score manually, e.g. if novice, ~500, if int~ 1000 if p/c ~ 1500 or whatever the system equilibrates to.
I think this will cover the main current problems, which are:
1) inflation of grades for people who have only ever had comps recorded where they win
2) bizarre results in secluded circles who mainly compete against each other, e.g. a group of beginners or so...
Robin
|
|
Joined on 14 Feb 2001
Total posts: 1,972
|
Couple ranking system
Tuesday, 16 March 2004 18:40
Two new pages upload for experiments!
When you go to any results, the page DisplayResults.aspx is displayed - if you change the address (in address bar) to DisplayResultsRating.aspx (same web site, same parameters - just insert word Rating in address) - you should see 2 additional columns - rating of couple before and after the competition.
If you click on the couple in this page you will see history of their results with ranking assigned after each comp.
We will reset all results tonight to experiment with new algorithm - so do not be surprised if you get 0 somwhere :)
As for initial score - the problem is that we have to start from something at the beginning, hence average would be ... null for first competition. Later on I agree average of ratings of people in the same round sounds as a very good idea. A bit of problem with encoding it, but it would be a challenge :)
Cut-off is implemented already (in the latest version).
I do not think that arbitrary assigning initial values is right. First of all 'do not mess up with self-developing system' :) - but more seriously - for most of events from around the world we cannot say without some investigation - what grade they are - just hope that these people will compete with others and that rating will straighten by itself.
|
|
Joined on 12 Mar 2004
Total posts: 2
|
Couple ranking system
Tuesday, 16 March 2004 20:24
Hi everyone. I have been reading the developing discussion with interest, not as a 'lurker', but as someone who has not been able to log in until now to join the debate. however, now I'm sorted, this will be, by necessity a post of some length as there are several comments i would like to add.
The topic heading that this debate is accessed from is not applicable to the subject matter anymore. Would it be possiblr to retitle it to attract a widwer audience to the discussion?
The idea of some form of ranking or ratins for Sunday circuit dancers below the EADA national selection grade would be a brilliant idea, but would have to be done with the blessing of the dance promoters association (DPA) and eADA in order to obtain comp results. Results from most comps are actually sent to Dance News and are eventually printed in an abbreviated form. Scrutineers,of course, collect the results and could, if agreed by all interested parties,post them promptly to a web site.
I have for a long time been using a semi-qualitative system of scoring to monitor our own dance performances over time. This uses factors to adjust the result in order to allow for :-
|
|
Joined on 12 Mar 2004
Total posts: 2
|
Couple ranking system
Tuesday, 16 March 2004 20:50
(cont)
- the number of rounds danced
- the position reached in the final (if any)
- the calibre of other competitors
- the status of the competition
In order to gauge a result I use points similar to EADA, ie 12 for 1st, 11 for 2nd, 10 for 3rd, etc, 5 for semi, 4 for quarter final, 3, 2, 1 etc for earlier rounds. However, I do this on a cumulative basis so that coming third in a comp when dancing a semi and a quarter final would give 10 + 5 + 4 = 19 points. If only the semi was reached in the same event only 5 + 4 = 9 points would be gained.
To obtain a result factor I simply divide our result points by those gained by the winner + 1. So, in the examples quoted above, this would give 19/22 = 0.864 and 9/22 = 0.409 respectively.
In order to adjust for the strength of the opposition I look at the status of the top 5 couples in the final. Each couple who are of the same grade as the event are given 3 points, couples of one grade lower are given 2 points and couples from two grades lower are given 1 point. The total is then divided by the maximum possible (15) to get an event strength factor. I then take the square root of this to limit its effect and multiply the result factor by it. Eg. if a senior amateur final included three senior amateur couples, one senior - pre and one senior intermediate the event strength factor would be the square root of 12/15 = 0.894.
I then use a competition status factor to add weight to big events like 'Stars' or 'Champions' and reduce results from very small ones. I base this on the number of adjudicators at the comp, using 7 as an optimum. The formula I use is not based on any scientific logic but has the desired effect. It is 1-((1/7)x(1/n)) where n = the number of adjudicators. Thus, a comp with 5 judges would generate a staus factor of 0.971 whilst one with 11 judges would give 0.987.
The status factor is then multiplied by the product of the result factor and the event strength factor and the result multiplied by 1000 to generate a normalised performance score.
The method above would probably be unsuitable fpr ranking couples, but it proves to be an effective way of monitoring personal performance progress.
|
|
Joined on 08 Jul 2002
Total posts: 20
|
Couple ranking system
Tuesday, 16 March 2004 21:13
As requested, I've renamed the topic. I also made it sticky; not that many other topics are crowding it...
Thanks for all the help guys - if there were more people like you this site would be twice what it is now
|
|
Joined on 14 Feb 2001
Total posts: 1,972
|
Couple ranking system
Tuesday, 16 March 2004 21:18
quite interesting and rather complicated system. Unfortunately totaly impractical for us - we just do not get all neccessary information to calculates all factors. Also - we are alloking for a system, which would be working cross countries and cross classes. As I have stated in one of the previous posts - we enter on average 5-6 events daily from around the world. It is simply impractical to enter event weight for each of them.
And yes - you are right - all systems discussed here are not for ranking, but for rating (i.e. assigning some value to the couple).
As for results directly from organisers: Guys - every week or two somebody comes up with this idea :) We would be more than happy if we would receive info from organisers of scrutineers. PLEASE - try it for yourself - help us.
On the next competition get results and send them to us. We promise to enter them almost instantly. You can send results in almost any electronic form (as long as it is not propriety binary format :)) - we have tools to convert or ... we can write one.
For obvious reasons I will not mention names - but you will not believe how many times we have had results promised and ... in the best case we got printed copy from one of the competitors. There was one exception 'South Of England' organized by Mary Richardson, where we got results in binary directly from scrutineers computer.
|
|
Joined on 14 Feb 2001
Total posts: 1,972
|
Couple ranking system
Wednesday, 17 March 2004 21:27
To those who are inteersted in progress :)
I think I have found one of the problems. Unfortunately our database contains some errors. Sometimes very good couple does not compete, but ... it is marked. We had once case when Bryan Watson had 4 marks in the first round! Obviously it is judges' error - but as for Bryan we know it must have been a mistake, in case of others we do not.
Also on competition like UK Open or British Open some couples are starting their competition from 2nd or 3rd round. If they do not turn up - they (wrongly) stay there - again: we cannot check it.
If such couple is in the first round - obviously everybody who 'won' with them has their rating increase ridiculously (If you won with Bryan Watson - obviously your rating should go very high!). I am not particulary good at chess (my rating is below 1400), but once I have won with 1700 guy (playing blitz - I was lucky :)) and my rating jumped high :)
Several people like that is starting to cause ripple effect on others.
That was explanation - now help me to find a solution :). One pbvious test comes to mind - if somebody's rating increased by more than let say 1000 - something is wrong. But what to do???
|