Improve ELO ranking

[Pistolero]

I'm thinking about the top players at 7Wonders who basically don't play ELO rank games anymore (It's probably the case for other games too). I think it doesn't reflect the actual competition. Once you reach the top, you disable ELO rank based forever to stay at the top level. I think all players should lose a very small amount of ELO points every day (like 1 per day, and only if you don't play ELO enabled games), so that they are forced to challenge themselves regularly in real competition.

[RicardoRix]

have a read:
https://boardgamearena.com/#!forum/view ... =9&t=13439

[Silene]

I already put this in the other thread - but since you started a general discussion about ELO-ranking here, i think this is the better place:


I think that playing multiplayer- game should have much less effect on your elo in general. I don't think it should be calculated in the same way as 3 single games. Maybe it is already reduced a bit, but I think it should be more like:

Devide the calculated ELO-change by the number of your opponents!
So I think if there's already some reduction in place (maybe that's why the numbers do not add up?) - it's just not enough.

The reasons:

• It is out of your control wether your opponents target you as an enemy or play in favour of you. That means there is more of luck involved than in 1 on 1. I think usually luck-heavier games lead to a lower elo-change in general (correct me if I'm wrong). This should be reflected in 3+ player games also.
• you played only 1 game after all. Should lead to similar ELO-variation as a 1 on 1.

For team games, there should be individual solutions. Winning as survivors in Not Alone is not a tie with your teammates. And the creature did not lose to 4 other players but lost to one other team (with elo = average of the 4 players).
So maybe devide not by number of players but by number of opposing parties and do not calculate anything against your teammates.

[might cause problems especially on Not Alone because elo-change is not symmetrical then. this would be ok if the propability for survivors to win is the same as for the beast. I think that's probably not the case so this would cause elo-inflation or deflation. maybe some factor between beast/survivor elo-change is in need, to balance that]

(fun fact: on solo-games you'll have to devide by zero. That's why thermophyles-elo is bugged )

[RicardoRix]

So you're saying if I come 1st out of a 500 player game then that's the same as just beating 1 other person in a 1v1 game?

How do you calculate that anyway? What is your new proposal for a calculation?
Should 1st place get the same ELO points as 499th place?

It is out of your control wether your opponents target you as an enemy or play in favour of you. That means there is more of luck involved than in 1 on 1. I think usually luck-heavier games lead to a lower elo-change in general (correct me if I'm wrong). This should be reflected in 3+ player games also.

This statement is not true.
If you keep the current system, then everyone fights for placements more, which means they're are not going to target anyone just for the hell of it.

If you play a game with more players, then you should expect to be rewarded for more time and effort put in.

@Pistolero
ELO is a mark of your ability, not the frequency in which you play.

[PlayFastOrByeBye]

I totally agree with you pistolero, elo is a ranking system, and should be living. Badges are here to show your abilities.

[Pistolero]

Once thing that is also extremely broken, for 7Wonders, is that I've done some games where it ends up (3 players) something like:

1 - 48 points, 24 gold
2 - 48 points, 23 gold,
3 - 48 points, 22 gold.

So, for 2 gold more or less you win or you lose (that point only is ok), but you get then either a +10 or a -22 ELO, which is complete non-sense.
All games very tight should reward like +1, 0, -1.

I think the discrepancy in points between the first and last player should be taken into account in the ELO formula.

Good players will try to sacrifice points so that their prime opponent will get even less points. But by doing so, they might end up 3rd, generally because the 3rd player is selfish. By taking into account the discrepancy, the punition is less worse.

[Jest Phulin]

I think the discrepancy in points between the first and last player should be taken into account in the ELO formula.

How would it be taken into account? What specific formula are you suggesting?

Winning a football (soccer) game by 2 points is a blowout win. Winning a basketball game by 2 points is about as close as you can get. Would the formula have to be adjusted for every game?
Also, would this be based on a percentage, or absolute amount? Is winning 1-0 winning by one point, or by an infinite amount?
What about conceding games like Reversi? That game has massive point fluctuations during the game. If a player is in a losing the game, will conceding at the beginning of their turn give a bigger ELO shift than if they concede at the end of their turn, when they'll have more points?
What happens if someone currently has a higher score, but can tell they are in a losing position, how will the points discrepancy be handled, since the winner has fewer points than the loser?
How are games that have no score handled? If someone takes all of the opponent's pieces in chess before they take the king, is that more of a win than someone who forces their opponent to box themselves in with their own pieces?

Yes, the current ELO system has problems. (The primary one being that people are using it as the victory points in the game of BGA.com instead of the tool to find opponents of similar skill level, but that's a different thread. ) But, every patch to fix one aspect of it that I've seen breaks several other aspects.

[Pistolero]

Since a victory 3-0 seems stronger than 53-50, the formula needs to take into account the discrepancy AND the total number of points (3 vs 103) which allows you to find if the discrepancy is huge or not, relative to the total score. This is quite simple.
Each game should have a default discrepancy, computed on statistics. Football is like 1, basketball is like 10, this is computable easily.
If you have a forfeit, you apply the default discrepancy.
For chess, ELO works fine, so you have your answer
I could easily find a nice formula, it's just pointless as long as the administrators don't agree to build on those principles

[ollyfish2002]

I could easily find a nice formula, it's just pointless as long as the administrators don't agree to build on those principles

Maybe we should just ask for a BGA-ELO that will run alongside to normal ELO.

[RicardoRix]

Yes it's tough and frustrating when you just, just lose. But in princliple this should happen as often as you just, just win, so hopefully that balances things out.

Interestingly and mildly relevant is that AlphaGo was interpreted to make moves that most likely led to a win even if other moves existed that would potentially lead to a larger margin win. Given how humans usually always go with bigger is better this style of play seems counter intuitive.

What is the best, to play for higher winning chance, or to play for a largest margin win?
You can argue both ends.

You may also find some strange strategies that have a lower chance of winning but if so, a BIG chance of netting a large margin victory and this becomes the new 'expert' (ELO grabbing) way of playing the game.