About Glicko:
https://en.wikipedia.org/wiki/Glicko_rating_system
How the rating system works
Sort:
Aug 9, 2018
0
#42
Aug 9, 2018
0
#43
"During the first games rating jumps are higher. Then they become milder." this is generally true, HOWEVER in 4pc all players have started with a very low deviation of 80. (we considered this the safest option as we unfortunately have no records how many games were played in the old version).
If you play opponets with similar rating to your own, deviation will got down to about 65, but not lower. if you play opponets with a rating that is far from your own, deviation might actually go up a bit from 80.
deviation = confidence that the rating is accurate. the higher your deviation, the bigger rating changes you will have.
FYI live chess also uses glicko, new players get a deviation of 350
Could you please explain this bug?
Aug 10, 2018
0
#46
Sep 4, 2018
0
#47
(This is a copy/paste from other topic. An answer to a Top Teams player)
> Glicko is probably fine, but how demoralizing it is to win 0.2 points and know that once I lose it will cost 50.
I understand you. But please look at these games of GM Hikaru Nakamura:
As you can see, he won a game against a 2424 player, but his rating is 3045, as it was before the game. So, he also got like +0.2. (The difference of their ratings is 621.)
This is your game:
You played against 1600+ players with a 1800+ teammate, so the difference is 1000+.
Please notice that you got +0.2 when your lower rated teammate got +0.1. Looks strange, at the first look. This is because it was your first game and your rating is unstable. Your teammate played 1020 games already, so ... Glicko is sure his rating is stable.
(Copied from this thread, as requested.)
GDII wrote:
The difference between Elo and Glicko is only the measure of rating uncertainty (RD). Other than that, both systems are pretty similar. The problem with Elo is that you can gain more points than you arguably deserve by playing against players with an unreliable rating. For example, you could be playing against players who just started to play or against players who haven't played in a while or against players who have an inflated rating because they played against other players with an unreliable rating. Suppose your opponent is rated 1500, but his actual strength is 1300, due to one of the aforementioned reasons. You would still gain the same amount of points as if you had played against a real 1500 rated player with accurate rating. As a result, your rating will be inflated. The Glicko rating system takes this into account by basing the rating change on the uncertainties of the players' ratings. So, it won't be as easy to obtain (or maintain) an inflated rating compared to the Elo system.
GDII wrote:
Similar to the Elo system, Glicko has a "K-factor" — although not called that way — which determines the maximum amount of points you can gain or lose per game. The difference is that for Elo this factor is fixed (e.g. 30), whereas for Glicko this "K-factor" depends on your ratings deviation (RD). If your deviation is high, your "K-factor" will be higher. So, if you don't want to lose a lot of points you should play more, such that your RD goes down. If you play less because you are afraid to lose points, your RD will increase and you will lose even more points.
To give an indication of how RD changes the K-factor (approximate values):
- if RD = 90, then K = 45
- if RD = 80, then K = 36
- if RD = 70, then K = 28
- if RD = 60, then K = 20
- if RD = 50, then K = 14
Conclusion: play more, not less
GDII wrote:
To make it even more clear: the Elo system is a special case of the Glicko system. The rating update formula is essentially the same:
R = R0 + K * (S - E)
where
R = new rating, R0 = old rating, K = max gain / loss, S = actual outcome (win 1, draw 0.5, loss 0), E = expected outcome (probability of winning, value between 0 and 1)
In the Glicko system, the K-factor and the expected outcome E are functions of the RD. However, there is a value for the RD that makes both systems identical. If the RD is approximately 73, then the K-factor will be around 30 and the rating change will be exactly the same as in the Elo system with K-factor 30.
If your RD is higher, you gain or lose more points compared to the Elo system and if it's lower, you gain or lose fewer points compared to the Elo system.
The problem: it has become too hard to gain a higher rating in both FFA and teams.
The solution: everything should stay the same aside from 2 details in FFA and 1 detail in team
Main idea: Rating caps (Once deviation has reached a certain level)
FFA
1: Cap for last place: a maximum rating loss should be set to 15, since finishing last often only means every one just ganged on you and doesnt mean anything about your play.
2: Cap for second place: A minimum of 5 rating points should be added regardless of the relative ratings of the players for the player who finishes in 2nd place. It makes no sens for second place to lose rating.
NB: in order to prevent rating inflation: any points added to a second place finisher or to a last place finisher should be substracted equallly among the other players. ie: I finishh last agsint weaker opponents and shoulld have lost 21 rating but only lose 15: the other 3 players each gain 2 points less than they would have. Or if a player finishes second and wins 5 points instead of ganing 2 points, all 3 other players make 1 point less. This way, the total rating changes in a given game stays the same. the only difference is that players who tend to often finish first second would be higher rated, and finishing last against lower rated opponents wouldnt punish as much. Hence, this should help having ratings that reflect playing strenght more accurately.
TEAM
1 Loss cap: a maximum rating loss of 15 should be set independant of rating.
NB. a maximum rating gain could also be set to 15, although this will not affect the top of the leaderboard since strong players will never get the opportunity to play for maximum rating gain. So I think it's better to not set a rating gain cap, as to allow new strong players to climb the rating latter quicker.
Setting a fixed gain/loss cap defeats the purpose of using Glicko. That's basically the same as reverting to Elo.
I think the only problem right now is that your deviation increases when your rating gets higher, because then you have to play against lower rated opponents more often than not. With a higher RD, you lose more points and normally you would also gain more points when playing equal opponents, but if you only get to play lower rated opponents, there's not much to gain — only more to lose. So, higher rated players have some (arguably unfair) disadvantage. The system (the RD specifically) is a little biased towards lower rated players.
In this game I only got 9 Points!???????????????????????
https://www.chess.com/4-player-chess?g=4642356
The rating system has been updated.
UPDATE AUG 18
Now we use Glicko system instead of ELO. Rating changes depend on amount of games played by a player. During the first games rating jumps are higher. Then they become milder.
Important consequence: Even if both teammates have the same rating, their rating change may be different.