CFC rating system broken?

[Vlad Drkulec]

Re: CFC rating system broken? [over/under2200]

It might seem that overall this is a wash, since every player whose rating increases from under 2200 to over 2200 gets to keep all of the points earned, rather than all the point up to 2200 and only half of those once over 2200. There's probably a clearer way to put that, but here's an illustration: a 2190 player whose tournment perfomance earns 40 points has his rating go to 2230, not 2215 as it would be if all the points after the first 10 (which brought the player up to 2200) were cut in half.

But for this to really be a wash would depend on how far above 2200 the player's rating had jumped.
Here's an example:
Player A is 2190 and earns 12 points to end up with a rating of 2202.
Player B is 2190 and earns 80 points to end up with a rating of 2270.

[Player A is only 1 point ahead of where he would have been if the lower K value was applied to all rating gains over 2200, while Player B is 35 point ahead of where he would have been.]

Both players compete in several events, scoring the same performance rating in each, raising their ratings each time, so that Player A is rated 2220 and B is rated 2285. Then they each lose a one game match vs a 2000 player.
The way the CFC ratings are supposed to work:

A's rating should drop 12.4 points to 2208,
B's rating should drop 13.7 to 2272.

The way the CFC ratings actually do work:

B's rating does go to 2272, :o
A's rating drops to 24.8 points to 2195.

Obviously this is not "a wash" for Player A, who benefitted 1 whole point in his step over 2200, but dropped 24.8 points on his way down.

Suggestion: it might be interesting to compare the number of active players between say 2199-2170 with the number between 2200-2229 to see if it's the Bell-shaped drop off you'd expect, or if there are a lot more active players just below 2200. If so, that would show a systematic effect of this rating error.

I don't have any suggestion about fixing this apart from rewriting the rating program and copying whatever wins this competition: http://www.kaggle.com/chess

Actually I haven't noticed that there is a problem on the way up through 2200. It seems to me that the calculation is done correctly and by the rules. The drop back downwards is calculated incorrectly and is a pain in the lower regions of his backside for Gerry who has to go through a manual procedure to fix it.

When your rating drops you drop a bit more than you should have but then things get a little confused if your rating goes back up because you gain back half the points that you previously lost due to the doubling of your gains under 2200. Also if you play a lot then the effect of a ten point drop gets further diluted by the fact that you gain two extra points by playing in a five round tournament if your rating is ten points lower (while you are under 2200). So in effect you get five of the ten points back when you have a good result and some of the points back just by playing because of the differential gains for being ten points lower.

[Vlad Drkulec]

Re: CFC rating system broken? [over/under2200]

In Bill's case the "loss or gain cut in half" would only apply against the 1st 12 points, taking him to 2200, after that the loss would be at the normal rate.

Ditto for you.

That is incorrect, at least with to respect to how the rule is written. The software does seem to apply the rule in the manner that you suggest but it is contrary to the way the rule is worded.

[Fred McKim]

Re: CFC rating system broken? [over/under2200]

That is incorrect, at least with to respect to how the rule is written. The software does seem to apply the rule in the manner that you suggest but it is contrary to the way the rule is worded.

Good point. I think the corresponding part, which I was referring to in the Doubleday illustration, has been omitted from the rules as presented.

So I'm suggesting the following should be added:

For players who start an event above 2199 and then in the event go below 2200 the loss is computed normally, namely with 16 in 714b and then the decrease under 2200 is doubled.

[Gary Ruben]

Re: CFC rating system broken? [over/under2200]

Good point. I think the corresponding part, which I was referring to in the Doubleday illustration, has been omitted from the rules as presented.

So I'm suggesting the following should be added:

For players who start an event above 2199 and then in the event go below 2200 the loss is computed normally, namely with 16 in 714b and then the decrease under 2200 is doubled.

Use the same K factor to the 2400 level. It should help the rating system to correct itself without interventions.

[Vlad Drkulec]

Re: CFC rating system broken? [over/under2200]

Good point. I think the corresponding part, which I was referring to in the Doubleday illustration, has been omitted from the rules as presented.

So I'm suggesting the following should be added:

For players who start an event above 2199 and then in the event go below 2200 the loss is computed normally, namely with 16 in 714b and then the decrease under 2200 is doubled.

You guys really want to make it unpleasant for people who care about their ratings. The net effect either way is somewhat negligible by the time you play a few tournaments.

[Denton Cockburn]

Re: CFC rating system broken? [over/under2200]

1) You can't rely on the published performance ratings as they are the "raw" performance that is used for unrated/provisional calculation. For example if you beat a 1400 player your pr for that one game is only 1800.

You have to go through game by game and do the math. Without using a calculator, I got about -87 points. Use 12 of those to to bring him from 2206 down to 2200 and then subtract 75 and he would be 2125, add 8 points for participation and you get 2133. Pretty close.

2) At the moment any changes done to the code, are by Vincent Chow on a contract basis. Removing the participation points was relatively simple, totally redesigning the code would be a major capital expense.

I'm a computer programmer. If the code is available, then I can fix this...for free too :)

[Egidijus Zeromskis]

Re: CFC rating system broken? [over/under2200]

For players who start an event above 2199 and then in the event go below 2200 the loss is computed normally, namely with 16 in 714b and then the decrease under 2200 is doubled.

I think, that after the person reaches certain rating (2000, 2200, 2400, whatever), the system should transfer a player to a "normal" category with coefficient 16 permanently. And treat the event as a whole, not devide into parts <2200 and >2199.

The rating system should be as simple as possible. It is better to have 1000 more players, than one perfect rating system :D

The CFC weekly rating updates really are publishing tournament results in timely manner. :p

[Bob Gillanders]

Re: CFC rating system broken? [over/under2200]

Good point. I think the corresponding part, which I was referring to in the Doubleday illustration, has been omitted from the rules as presented.

So I'm suggesting the following should be added:

For players who start an event above 2199 and then in the event go below 2200 the loss is computed normally, namely with 16 in 714b and then the decrease under 2200 is doubled.

Okay, so I have worked thru all the calculations myself and I am satisfied that the system is working fine. As Fred has described, the rating program is using the appropriate k factor on a game by game basis. Sorry John and Bill, once you dropped below 2200, the losses doubled. :(

I will send you my worksheet for your piece of mind. :)

Fred - I will advise Kerry to make the Handbook corrections. :)

[Alan Baljeu]

Re: CFC rating system broken?

Words tend to be imprecise. We should stop defining ratings with verbal requirements like Fred's and go with clear algebra. Use a simple algorithm and remove all doubt of interpretation. Write examples of input and output to demonstrate it works. THAT is how the rating rules should be defined.