Universal Rating System

Discuss anything you like about chess related matters in this forum.
Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Thu Jan 05, 2017 9:37 pm

Lewis Martin wrote: So the gaps do not make any sense to me.
On the chess.com site, there is a theory that they represent the loss of skill if you matched a player playing at G/120 or similar against one playing at G/30 or similar for rapid or G/5 or similar for blitz. Somewhat of a pointless statistic if a correct theory. Someone like Jack, playing all his games at rapid speed should get an Rgap of zero on that basis.

The point has been made, probably correctly, that "long" ratings are just as good predictors of Blitz performance as Blitz ratings. I don't see that as a justification for modifying "long" ratings to reflect Blitz performance. But there's insufficient data. What ratings would you get if you ran the "Universal" method against only standard play results?

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Thu Jan 05, 2017 9:48 pm

E Michael White wrote: Wonder if they've assessed whether under normal load it contains internal inflation/deflation, spread/contraction and whether these are uniform over the whole range of grades.
Isn't the underlying premise a more sophisticated version of the ECF's new player routine running for a 72 month period? That type of approach is always able to produce relative measures of strength and I would suspect the gap between highest and lowest would change only if the highest player was able to outperform the rest. That said, how would they anchor it? Two problems - one is that you may not want the mean or median creeping up or creeping down unless the rated population changes (which it will in an International context) and the other is that the relative standard between top human and lowest rated human may widen over time if the top human gets better.

Aside from the sophisticated points, it has to answer the charge that there's some false data in there as witnessed by the discovery of inactive titled players with near beginner ratings.

MartinCarpenter
Posts: 3043
Joined: Tue May 24, 2011 10:58 am

Re: Universal Rating System

Post by MartinCarpenter » Fri Jan 06, 2017 9:24 am

Roger de Coverly wrote:The point has been made, probably correctly, that "long" ratings are just as good predictors of Blitz performance as Blitz ratings.
Really? I'd expect the blitz ratings to work rather better in principle - some people definitely have distinctly different performances at the two time limits. Maybe if there isn't enough data in the blitz ratings or something.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Fri Jan 06, 2017 9:43 am

MartinCarpenter wrote: Maybe if there isn't enough data in the blitz ratings or something.
It was a claim made by Sonas about the recent world Blitz in support of the notion that there's a central chess "ability" invariant across all forms of chess. That might be true, but to the extent that if you lose on time in a decent position through not playing quickly enough, that's the result the ratings will report. With increments, losing on time is less of a risk.

The lack of detail means it's speculation as to the underlying principles of the URS method, but there appears to be a notion of an underlying chess strength which is devalued by having to play at faster time controls. Two players of identical underlying chess strength may have different devaluation factors, therefore the predicted outcomes for Rapid and Blitz differ from those for Classical.

Elo also had the concept of an underlying chess strength but recognised that actual game results gave an unreliable measure of what it was.

Kevin Thurlow
Posts: 5820
Joined: Wed Apr 30, 2008 12:28 pm

Re: Universal Rating System

Post by Kevin Thurlow » Fri Jan 06, 2017 9:50 am

"He added that a purely random ratings system would only predict the winner of a decisive game 50 percent of the time, while a perfectly accurate system would predict the winner of a decisive game 100 percent of the time. His goal for URS is "75-80 percent," which is better than current prediction systems according to him."

"A perfectly accurate system" is a nice idea, but as likely as a perpetual motion machine. All rating systems provide a measure of past performance, that is all. If people's "standard play" rating is affected by "blitz", then they might stop playing rated blitz events. Blitz ratings are always going to be fairly random as blitz chess is fairly random.

The URS is interesting, although obviously they have to iron out some problems.

There is a big advantage (for the supplier) in having incomprehensible systems of calculation - it makes it very difficult for players to complain if they think ratings are wrong.

User avatar
Paolo Casaschi
Posts: 1186
Joined: Thu Jan 08, 2009 6:46 am

Re: Universal Rating System

Post by Paolo Casaschi » Fri Jan 06, 2017 9:55 am

Angus French wrote:So... the URS has in fact used all the results of all the games rated by FIDE in the last 6 years (and there are 246,022 players in the initial URS list - as I found after downloading a copy). And presumably this data has been willingly provided - for a fee? - by FIDE.
The data is available for free from the FIDE website, you can download all of that yourself if you want to.

It's not readily available for download in one click, but it's all there and not to difficult to have an automated script doing the job for you.

For example:
  • this is the crosstable for the e2e4 Brighton Open, the first event in the page above; it contains all results for each player, even the colors; repeat this for all rated tournaments and you have all the data you need
No need to pay ant fee to FIDE or even to ask for permission; the info seems just publicly available; I did not spend much time looking but I could not find on the FIDE site any copyright claims for the data or any acceptable use policy of any sort.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Fri Jan 06, 2017 11:06 am

Kevin Thurlow wrote: "A perfectly accurate system" is a nice idea, but as likely as a perpetual motion machine.
Rating systems can predict results in the sense that between players around 200 points apart, you get 75% to the higher rated player. What they don't predict and the newer URS system is no exception, is what the drawing percentage is and what is the percentage where the result goes the wrong way. Wrong way results where the lower rated player wins make a nonsense of the claim that 100% reliability of the prediction of a decisive game is even possible.

I think I've figured why they think a universal rating is possible. In effect they propose different (lower) rating scales for Rapid and Blitz where the conversion between the two is player dependent. The default values being 32 and 84. So if you play a rapid game, that counts as a "Classic" game against someone plus or minus the respective Rgaps. In theory at least, this may iron out the variability in performance over different time controls. Perhaps the concept should be presented as the abolition of separate rapid and blitz lists, but with player dependent factors to establish the rankings for these forms of chess.

Angus French
Posts: 2151
Joined: Thu May 15, 2008 1:37 am
Contact:

Re: Universal Rating System

Post by Angus French » Fri Jan 06, 2017 11:14 am

Paolo Casaschi wrote:
Angus French wrote:So... the URS has in fact used all the results of all the games rated by FIDE in the last 6 years (and there are 246,022 players in the initial URS list - as I found after downloading a copy). And presumably this data has been willingly provided - for a fee? - by FIDE.
The data is available for free from the FIDE website, you can download all of that yourself if you want to.

It's not readily available for download in one click, but it's all there and not to difficult to have an automated script doing the job for you.

For example:
  • this is the crosstable for the e2e4 Brighton Open, the first event in the page above; it contains all results for each player, even the colors; repeat this for all rated tournaments and you have all the data you need
No need to pay ant fee to FIDE or even to ask for permission; the info seems just publicly available; I did not spend much time looking but I could not find on the FIDE site any copyright claims for the data or any acceptable use policy of any sort.
I wondered if a scraper app had been written. *If* that's what they've done then FIDE could at any time easily prevent access... How ever the data has been obtained, the URS is reliant on the cooperation of FIDE and, unless there is a contractual agreement, that presumably constitutes a rather significant risk.

Stewart Reuben
Posts: 4542
Joined: Tue Apr 03, 2007 11:04 pm
Location: writer

Re: Universal Rating System

Post by Stewart Reuben » Sun Jan 08, 2017 1:54 pm

Th London Chess Association, had its own grading system when run by Tony Gaffney. I think it was in the 1980s. Roger would be able to pinpoint the period. Basically
rapidplay k factor = 1 fold.
quickplay finish or adjudicated 2 fold
games played to a finish (e.g. London League or Lloyds Bank Masters) 3 fold. I probably have some details wrong.
The list was published monthly. It was used to define eligibility for grading restrited events,
That was at my suggestion. It continued for a few years and helped popularise London competitive chess. It had nothing to do with the BCF. It fell into disuse when Tony lost interest.

It was at my suggestion FIDE extended its list down to 1000. I don't know whether it has been considered extending into down to 500 or even 0. Of course there would be some problems. Having 3 different lists for standardplay, rapidplay and blitz seems robust enough currently. I don't know how online and integrating that is going.

Richard Bates
Posts: 3338
Joined: Fri Nov 14, 2008 8:27 pm

Re: Universal Rating System

Post by Richard Bates » Sun Jan 08, 2017 2:17 pm

Stewart Reuben wrote: Having 3 different lists for standardplay, rapidplay and blitz seems robust enough currently.
It's not clear exactly what you mean by "robust".

London Chess Classic Super Rapidplay prizewinners:

Under 1600 Rating
R Praggnanandhaa IND +5.70 £250
Sharma Dushyant IND +5.68 £150
Nagpal Vardan IND +5.20 £75 Junior

FIDE rapidplay (and blitz ratings) are at the moment still clearly a work in progress. The fundamental problem they have is that they are originally derived from an individual's (classical) FIDE rating (thereby making an assumption that one is a reasonable basis for another) but from then on act completely independently. Because many people don't play much rapid chess the two can therefore diverge quite significantly for no good reason. FIDE could combat this artificially setting a maximum on this difference to say, 100 points, but then this would create an alternate problem for those who do actually play significantly more in rapid events (which may be particularly true in England) if large changes in strength evidenced in these events were restricted by a static classical rating. The alternative of allowing a limit to work both ways (ie. allowing large changes in rapid rating to move the classical rating) would probably not be acceptable, even if the most sensible solution.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Sun Jan 08, 2017 6:50 pm

Richard Bates wrote: The fundamental problem they have is that they are originally derived from an individual's (classical) FIDE rating (thereby making an assumption that one is a reasonable basis for another) but from then on act completely independently.
The URS method with Rapid and Blitz ratings are linked to the classical rating by a player dependent factor might have its pragmatic points. It distorts the ranking order for classical chess as witnessed by the higher rankings in the URS order for some players known to be relatively stronger at fast time limits.

Ranking and prize eligibility for the Classic Rapidplay might actually work better on the URS ratings, although not if some long retired British titled players show up on the basis of what appear fictional results.

Roger de Coverly
Posts: 21301
Joined: Tue Apr 15, 2008 2:51 pm

Re: Universal Rating System

Post by Roger de Coverly » Wed Feb 01, 2017 11:56 am

1st February update?

FIDE's list has appeared on its usual schedule.

David Sedgwick
Posts: 5249
Joined: Mon Apr 09, 2007 5:56 pm
Location: Croydon
Contact:

Re: Universal Rating System

Post by David Sedgwick » Wed Feb 01, 2017 9:16 pm

Roger de Coverly wrote:1st February update?
The February 2017 URS Ratings are now available at http://universalrating.com/ratings.php.

A press release about the February list should be available shortly.

Ian Thompson
Posts: 3551
Joined: Wed Jul 02, 2008 4:31 pm
Location: Awbridge, Hampshire

Re: Universal Rating System

Post by Ian Thompson » Wed Feb 01, 2017 10:52 pm

David Sedgwick wrote:The February 2017 URS Ratings are now available at http://universalrating.com/ratings.php.
... but may not be very accurate.

They only have 2 games recorded for me in January, when I actually played 6 games. That could mean my 2 4NCL games were rated, but my 4 Rilton Cup games weren't.

However, someone else who played 9 games in the Rilton Cup, and nothing else in January, has 5 games recorded in the URS list.

A third player at the Rilton Cup played 19 games in January in 3 events, but only has 13 games recorded in the URS list, and you can't get to 13 games by omitting whole events from the total because he played 2, 9 and 8 games in the 3 events. In the FIDE list he also has 8 games rated in January that should have been in the December list, but were presumably submitted late for rating. I've no idea how the URS rating system deals with late submission of events.

Leonard Barden
Posts: 1857
Joined: Wed Dec 24, 2008 11:21 am

Re: Universal Rating System

Post by Leonard Barden » Wed Feb 01, 2017 11:59 pm

David Sedgwick wrote:
The February 2017 URS Ratings are now available at http://universalrating.com/ratings.php.
I have to say that the February 2017 URS ratings I looked at are complete and utter crap.

I was interested in a junior age group and to see how the URS ratings compared with Fide.

I was shocked to see that the top three in the world on this particular list do not appear at all on Fide. So I tried to work out why, but the player histories on URS give no result information, just a list of alleged ratings from previous lists. In contrast Fide ratings, or for that matter ECF grades, provide the full list of individual games for the rated player, dating back several years.

Further, although tne URS ratings for Carlsen & Co have numbers you would expect, every checkable number on this junior list (which has around 30-40 names which do not correspond with the similar Fide list) is inflated by around 200 points, suggesting there is some serious generic fault in the software.

Trying to solve the mystery of the three unknowns at the top of the URS alleged global junior list, I entered each name in the Fide search.

Player 1, rated 2000+ by URS, has never played a Fide rated game.
Player 2, rated 2000+ by URS, has played two Fide-rated games, in each of which he managed to defeat an 1100 opponent.
Player 3, rated 2000+ by URS, has never played a Fide rated game.

It's really quite appalling that before this list went public that nobody made very basic checks on whether the URS computer was spouting nonsense.

The existence of this list is an insult to the many hard-working people who ensure that Fide ratings and ECF grades are comprehensive and accurate.

Post Reply