Forums Home / The Lobby / Expanded Pedro Seeding Post (
View Older Thread |
View Newer Thread)
First |
1 |
Last
| Q
- 12 Jun 2012
Total Posts: 497
|
goran said: | Putting the 192 points back to Pedro's rating will put Pedro at 13 right now. Just like Davis's list. |
First, I like the idea of having the system that Travis proposed put in place where instead of a 2 tiered classification system for rating status we have 3. (active, inactive, provisional/new...see Travis' post for more explanation) There has been precedence for this I believe in the chess community in previous years to which I am trying to gather more information on to make parallel assesments/apply to air hockey matches as well. I have been reading a lot since Tim and Don brought up the question of Pedro's seed/point change after the match to Daniel about the Elo system and its use in chess which it was originally designed for. It is of course used elsewhere now including video games, table tennis, foosball, etc. There is even use documented in chess of adjustment factors which parallel the "+1 rule" which has been the subject of debate here and when it was first added. These factors were eventually removed for various reasons. Other mathematicians have made adjustments to Elo's system in hopes to improve the goal of the system which is to have a statistical rating that approximates the strength of a player in order to draw out an expected percentage for the higher player to win any given match. In any procedure we can lay out, there will be anomalies and areas that come up where we can improve, but the main goal for our use in AH is and has been imo to use the ratings as a relative measure of success/skill in order to properly seed a tournament. With everyone's rating there is some standard amount of error...a player who plays one recorded match would have a rating with a high amount of uncertainty whereas a player who plays several matches against several players who have played several other players and so on would have a much lower uncertainty. A lot of this has been sidetracking from the point at hand which is adjusting Pedro's rating or applying the "L&L" rule as Don called it. Am I right that any such change will require a motion and vote by the board? In either case, I thought I would just lay out my 2 cents on the subject as well as a general question.
Before acting on this, I am curious are we treating this as just an isolated incident that needs fixing and then addressing the cause of the problem? Or are we hoping to modify the system as a whole to remedy the situation? I believe that we could rationalize the former for an immediate fix but a little bit of work and study should be done before completing the latter. (Any math majors who need a senior project out there? I've come across a few articles/thesi about Elo and it's application in chess...could do a case study on AH)
|
|
|
|
| Q
- 12 Jun 2012
Total Posts: 497
|
To start, in my opinion Pedro is coming in under-rated and therefore underseeded, but I want to approach the solution without bias and examine the cause if there is an issue. Arguments could be made about other people's seeds being slightly off but to approach any one particular case, need to look at the facts. I have only seen the few Youtube videos that Pedro posted and read about/heard everyone's talk about his previous tournament performances which is my initial basis for believing a rating of 1600 is too low for Pedro. If you look at the match against Daniel in December, that was the first match that Daniel had ever had reported on AHW. At that stage, Daniel is an unknown player of unknown skill. He had an initial rating of 1200 at this point but the uncertainty in his rating was at least 2-300 as a new player. Pedro, at the time was considered inactive which meant in our current system that his rating would be affected by playing Daniel. Arguments could be made that Pedro's rating at the moment he stopped playing AH in whatever year that was had an extremely low uncertainty since he had played against several other well established players. Even with the time off between the day he stopped playing regularly and the day he played Daniel would only increase the error in his rating slightly. Arguments in this case could be made that Pedro was the accurate rating benchmark on which to judge/better estimate Daniel's rating along with the other Venezulean players that made it to Vegas in 2011 (Javier and Dionzio). Daniel won 4-3 in the challenge match, with the math working out that there would be a 192 point rating increase for Daniel and because of the "2 classification system" a 192 point rating decrease for Pedro and this is the source of the error that a lot of players are complaining about. This shows how the movement to the "3 tiered system" would have helped. If that was in place, Pedro would have had his rating decreased at the time by 10% of the difference between his rating and 1200 at the time of being marked as inactive, and his rating would have been unpreturbed by the loss to Daniel. Daniels rating would then be up close to 1400 with a slightly lower amount of uncertainty than before. Truth is based on a very low number of matches overall, it is difficult to say with a high amount of certainty where Daniel should be rated. Anywhere from 1400-1600 given the data makes sense to me based on his opponents and his opponents'opponents. The other thing to note is that if Pedro's rating was unchanged because he was playing a new player of unknown "true skill" where does Daniel's rating fall. (1640 was due to the +1 rule because of Pedro's rating decline. 1400 would be the new rating for D.O. if the +1 rule did not apply due to Pedro's non changing rating)
This has been a long ramble on my part which is why I put it in another thread instead of fulling hijacking the "Pedro thread" in T&C. The thesis statement I am trying to get across is that there is mathematical justification for the solution that Goran has proposed, but I would be initially careful and cautious to make it a global change to the system on short notice and without a careful oversight for causes of other errors.
|
|
|
| goran
- 12 Jun 2012
Total Posts: 428
|
| For the +1 rule Daniel would be rated in the 1800s with Pedro. I like the idea of inactive players rating not to be effected by establishing players, and we don't need the. +1.
|
|
|
| TWeissman
- 12 Jun 2012
Total Posts: 210
|
I am very familiar with the ELO system. It is used in competitive card gaming. I ran high-end tournaments for 15 years in this area.
The bottom line is that there should not be a +1 rule, and there should be a vastly more-restricted maximum point win/loss.
Right now, the system of ratings in Air Hockey is near useless with the +1 rule. First, eliminate that ridiculousness.
Second, there needs to be a maximum of maybe a 64 point swing for a challenge match, 32 or 48 points for a World Championship match, and then on down the line.
|
|
|
|
| Q
- 12 Jun 2012
Total Posts: 497
|
TWeissman said:
Right now, the system of ratings in Air Hockey is near useless with the +1 rule. First, eliminate that ridiculousness.
Second, there needs to be a maximum of maybe a 64 point swing for a challenge match, 32 or 48 points for a World Championship match, and then on down the line.
|
Tim, I agree with you about the +1 rule for an improved rating system. Arguments were made (especially back just prior to when it was voted on by the USAA) about it being unnecessary and there are those that feel the same as you do.
As for the maximum point swing if I am following your post in the terminology of the Elo system, what you are suggesting is a k-value adjustment across the board? The maximum swing of a challenge match would be someone winning a 4/7 4-0 so each set of a challenge match would be a kvalue of 16? (unless you want to have it lower and add a CM bonus). Or are you suggesting leaving the 32 k-value, CM bonus, and just having a ceiling for change? Would you then put the k-value for a single 4/7, 3/5, 2/3, 1/1 rating set lower than 16? I do like that if the kvalue is lower and ratings are used for seeding, it would hopefully motivate people to play significantly more to try and improve their seeding.
Any changes at this point though, would require a board vote at this point if I understand policy correctly since it was voted in as the method for seeding the Houston tournament?
The situation at hand though and how to proceed is still regarding Pedro's rating. Do we look at just his rating/seed and apply the L&L rule as a one time fix or do we go back to 11/6/11 and recalculate the Venezuelan players' ratings (easy since only 8 players and about 50 sets/matches in the database since 11/6/11)? Are the fixes to the rating system needed before we can make these adjustments? Look forward to everyone's input.
|
|
|
|
| goran
- 12 Jun 2012
Total Posts: 428
|
| I agree with Tim on the +1 rule. The reason the k values are so high is to get an accurate adjustment for the little air hockey that's played. If there were more majors or even weeklies without point spots, I can see lowering the k value. The k value is lower for players over 2000. It might be a good idea to lower it to 1600-1800.
|
First |
1 |
Last
Forums Home / The Lobby / Expanded Pedro Seeding Post
|
|
