Generated Feb-25 21:37:10 GMT

You are not logged in.

Or: Recover Account (Forgotten Password) / Register new user / Re-send validation email / Main Page

**Maintenance is currently disabled.**

This thread is open.

Posted by **David Parr** 2012-07-10 19:48:10 GMT

If you want ratings that work then ratings should be calculated like the US Chess Federation does it.

1) Everybody starts with a 1600 rating.

2) In a 2 player game the winner gains (16 + ((loser's rating - winner's rating)/25 3) In a 3 player game 2nd place does not gain or lose and use above formula to calculate new ratings.

4) In a 4 player game calculate gain and lose using above formula for 1st vs 4th position and 1/2 the amount for 2nd vs 3rd position.

Posted by **magnateKit** 2012-07-12 16:33:00 GMT

This is a little too simple. For example, if one player has a much higher rating than all 3 other players and comes in 4th, the 2nd and 3rd place players would see no benefit from finishing ahead of that player in hti scenario. I think the formula would need to be tweaked to take in more detail. Each players rating should take into account the ratings and relative finishes among all 3 players.

Posted by **LAMBOLT** 2012-09-01 12:21:17 GMT

It would be interesting to see something done with your average score, perhaps in some kind of normalised z-index way so that the points you score are contextualised within the amount of points scored in the game you played

Posted by **Mol** 2013-05-25 07:23:34 GMT

I propose a variant on the above mentioned way of implementing the ELO rating. In a 4 th player game:. Player 1 wins from 2, 3 and 4. 2 wins from 3 and 4. 3 wins from 4. So you apply the the ELO rating 3 times for every player. For a 3 player game you apply the same principle. 2 player normal ELO rating. Is there a way to experiment with this way of calculating? To me it looks rather fair. Any comment?

Posted by **Renfield** 2013-05-25 15:58:07 GMT

I would suggest that a composite rating is flawed to begin with. Why not just simply create separate ratings for four player, three player and two player games? The difference between a two and four player game are considerable and should be viewed differently.

There should also be an effort made to collapse the impact of playing (and beating) the same person over and over. yzemaze, the current top rated player, simply beats up on Hagomir in two player matches. His last 20 matches are comprised of 18 2 player matches in which he is 16-2 against Hagomir and two three player matches against Hagomir and Tardor. At the best, yzemaze is playing uncompetitive matches, at the worst... In any event, I doubt that he is the best player on the site.

Posted by **Dude** 2013-06-19 15:29:57 GMT

Try this Prov = less than 12 s1 = 1 if A wins, -1 if A loses, and 0 for a stalemate Established = greather than or eqaul to 12 s2 = 1 if A wins, 0 if A loses, and 0.5 for a stalemate conditions (Ra*Na+((Ra+Rb)/2)+(100*S1))/(Na+1) 1 empty column leave alone Player 1500 Player b 1520 (Ra*Na+((Ra+Rb)/2)+(100*S1))/(Na+1) 1550 2 If A & B are provisional (i.e both less than 12) ((Ra*Na)+Rb+(200*S1))/(Na+1) Player A 1500 Player B 1500 1563.333333 3 If A is on provisional and B established Player A 1600 Ra+((32*Nb/10)*(S2-(1/(1+10^((Rb-Ra)/400))))) Player B 1516 1605.75896 4 If A is on established and B provisional 5 Player A 1600 Ra+(32*(S2-(1/(1+10^((Rb-Ra)/400))))) Player B 1516 If A & B established Great, maths! I just love it...

Posted by **Dude** 2013-06-19 15:32:30 GMT

I agree that the current system needs a overhaul as there are flaws and a system that you have a rank seperatley for ech 2 3 4 player would also work better I dont mind helping with this.

Posted by **SlothNinja** 2013-06-25 12:58:35 GMT

I suggest checking out the Glicko system (http://www.glicko.net/glicko.html). It gives each player a rating R and rating deviation RD. The RD is basically a confidence level in your rating R. Basically, the system is 90% sure that your true rating is better than R - 2*RD or 80% sure your that your true rating is R +/- 2*RD. One of the nice things is that it takes into account the strength of the players when updating the R and RD values. So, a strong player constantly beating the same weaker player is unlikely to raise their rating by much.

Posted by **Dvd Avins** 2013-07-08 21:57:19 GMT

I think Moi's suggestion is on the right track, but I don't think it gives adequate weight to winning the game. I think each match involving the winner should count twice as heavily as each match that does not include the winner. For example, suppose equally rated players in a four player game, with a k-factor such that each matchup not involving the winner is a bet of 10 points and each matchup involving the winner is 20 points.

The 4th Place finisher would effectively 0-4 (because the match against the winner counts twice) and would lose 40 points. The 3rd Place finisher would be 1-3 and would lose 20 points. The 2nd Place finisher would go 2-2 and would have no rating change. The inner would go 6-0 and would gain 60 points.

In a similar 3 player game, 3rd place would be 0-3 and would lose 30, 2nd Place would be 1-2 and would lose 10, and the winner would gain 30.

IN a 2 player match, the lose would give up 20 points and the winner would gain 20 points.

Posted by **Chrisr** 2013-07-23 04:08:15 GMT

Personally, I think the best rating system would mirror something like star craft 2.

1) the actual algorithm is secret to deter gaming the system. 2) players are only rated master-diamond-gold-silver-bronze. No actual number.

Posted by **Brent** 2013-08-15 23:24:08 GMT

Here's an idea for a formula that would work for 2,3,4 player games. It's 'Points Based', not wins based because we want to avoid king-making. Ie, if a person is coming in last, they should still salvage every single point they can instead of screwing people. Also, it should be worth more points by destroying the competition, and it should LOSE more points when getting destroyed.

W = Winners score M = Mean score (The AVERAGE score of all players in a game) S = Your score

A = Average rating (The AVERAGE rating of all players in a game) Y = Your rating

K* = average ratio of win from mean (a constant)

Rating Change = M*(((A/Y)*(S/W) - K)

*Suppose the average winning score in thousands of sampled 4 player games has 150 points, and the average score is 130 for all players. In this case, K = 1 + ((130/150) - 1)/2. K will be 0.933 for these examples.

Note: K could vary for 2,3 or 4 player games, based on statistics gained.

Note: Playing players significantly lower rated cannot gain points, even in a win (unless it was a substantial win). Therefore some tools might be needed for top players to find opponents who can improve their rating. Examples below will illustrate.

Note: This is a statistical NET ZERO system, so an initial starting rating is required. 1600, 2000, 3000, whatever could be a starting value. Choose a value that won't likely reach negative numbers for very poor players based on the average rating changes and lots of losses. Negative ratings could be possible.

Here's some examples. For sake of simplicity, assume A = 3000, M = 140 and W = 150 in these examples.

Higher Rated Player Wins: Y = 4000 S = 150 Rating Change = -26 (you are playing scrubs, your rating is way too high for the average. Try a 'fun' game instead if you can't find an opponent. No one should be impressed by this win.)

Higher Rated Player Loses: Y = 4000 S = 130 Rating Change = -40 (even worse)

Lower Rated Player Wins: Y = 2000 S = 150 Rating Change = +79

Lower Rated Player Loses: Y = 2000 S = 130 Rating Change = +51

Equal Rated Player Wins: Y = 3000 S = 150 Rating Change = +9

Equal Rated Player Loses: Y = 3000 S = 130 Rating Change = -9

The last 2 samples are what happens when closely matched players face each other. The cream will rise to the top, as is intended by this system. A scaling factor could be applied to each gain/loss of points to amplify the importance of the most recent results. So it could be +90/-90 if we want to accelerate separation. Also note that a player could NEVER win a game, and be the top rated player, so long as they are always very close to the winner (not very likely though).

Posted by **Edward** 2013-08-19 05:32:06 GMT

Brent's idea looks very good.

Posted by **Bogdan_J** 2015-02-23 19:50:32 GMT

Is that all? All of these ideas, discussions, and ... nothing? On the website is still the same old rating system, which speaks of ... nothing. Apart from the fact, that someone 5 months did not have time to play? Does the site administrator tried to change the rating system?

Posted by **Mol** 2015-03-08 11:28:22 GMT

The content of this message has been cleared by the person who posted it.

Posted by **Mol** 2015-03-08 11:43:03 GMT

I am a big fan of this site. Maybe the rating system is not perfect. But let us not complain, the administrator has done a great job.

Posted by **JMW** 2015-03-22 19:17:52 GMT

I'd agree with Mol. If we paid to play it might be different, but as it is I think we have to accept the current system for what it is (a fairly reasonable representation of your results over the past 5 months)

Click here to return to the Board Page, or here to return to the Main Page.