ELO Rating Improvement

[SonGoku]

The Problem of ELO rating

I have been thinking about players disappearing after 60 days from rating but they wont lose their ELO.

Well. the main problem is , though we got rid of forever banned and inactive players , that after some period players tends to climb up in rating even to top5 and then they will stop to play.. Before or after 60 days come , they will play 1 game. It means they discovered advantage of 60 days rule and.. its.. kinda bad, i explain .

Maybe some of you dont know , but ANY board , card , society (and similiar) game is constantly evolving itself by players.. When (never if) players like a game and they want to improve, it comes naturally they will , simply by playing it.. More interested players will even study a game , by watching replays of top ranked players and self-study by watching theirselves in replays, what they did wrong. This is how a game is evolving itself..

When we take those arguments into TIME aspect , it means when (for example) I was the best of 7000 players in Stone Age in year 2008 and i stopped playing , i would be probably this year only top50 , or even top100 place consisting of that same 7000 players. If we add more players , like the amount is now in Stone Age , i would probably be kicked off from top100 and go cry to the bed.

So back to the 60 days rule - After climbing up , top ranked players will play 1 game per 60 days , so actually ELO ranking becomes wrong at this point and TIME aspect. When they keep doing it , they will be still the first, because they also tend to find a low-ranked player to keep/go back into ELO rating , BUT their skill will be much lower than players between top10 - top50 trying to reach best places in rating , because those players strive to go higher and they have to play against superior opponents , which evolves them. Meanwhile top ranked players keep their ELO and skills also, not evolving.

In the end, this is clearly exploiting ELO rating advantage towards lower ranked players.. It doesnt achieves the spirit and matter of skills , rivalvy , improving and many other skills.

SOLUTION

I have seen kind of negating ELO score affecting any player on unnamed website providing Browser Card Game.. The only difference is that there are weekly tournaments and whole ELO rating resets each week.. It means i don't know how would it work for Overall non-reseting ELO rating.. But i guess it would go well.

The system run like :

Any player who didnt play a game after "determined" time , will start to lose ELO back to the default (in our meaning back to 1500).
They had weekly tournaments, so it means when i started to climb up Monday from 1500 to 1800 and stopped playing, in Thursday my ELO started to go down to 1500... My ELO score dropped after each day , but gradually.. Thursday was for example -20 , but Friday -30 , Saturday -40 and Sunday -50, then tournament reseted , gave away results of standing and rewards according to them and another tournament began..

How to apply it for our Overall ELO (non-reseting) Rating ? :

Somewhere between 20 - 30 days inactivity would fit to any game (which is constantly evolving) , then ELO would start to drop down each day and grow in negative points daily..

The only question is how many points we should apply to all players regardless to their ELO , but balanced points regarding to their ELO .

If you like my idea , start to discuss , what kind of improvement do you see and/or how many points should players lose after how many days ? (i gues 20-30 days is ideal).

Thank you for reading it ,

SonGoku

[cctsai]

I'd love to say I think SonGoku point out a pretty important issue, and hope some improvement toward this direction will appear in the future.

-

A proposal towards SonGoku's idea: I think it's probably comparably easier, for programming and for experiment purpose, to add an "inactivity penalty" in the following way: A player with good ELO points (say over 1700 maybe, gradually) gains inactivity penalty if she or he hasn't played with a number of different players of equal levels or higher for a long time. This inactivity penalty doesn't really affect the original ELO points (so the central algorithm for ELO system don't have to change), but our modified ELO points will be

Modified ELO points = Actual ELO points (following the original ELO system algorithm) - inactivity penalty.

And people are ranked according to these modified points. In addition, if it turns out unfortunately that this creates more new problems, then at least we can just turn it off and go back to normal.

Edited: another feature I think this could be good is that a player is not necessarily punished for not playing for a long time; as long as they come back to play and demonstrate equivalent skills, they can regain their ranks.

[Oso]

Ammm I prefer 45 days, and after this day, player will lose 2 ELO per each day. Really, when you've +1800 ELO points you don't get lot of ELO.

In BGA, try to get high ELO it's so HARD, in some games when you've +1600 you got a TOP 20 (like Uncle Chestnut, Northwest Passage, etc).

Generally people don't play more than 10 games at same time. In people like me (I played all games), it will be so hard because I should to play all games again.

[gdBlaze]

I'd like to think about what BGA wants to accomplish. Which is to prevent a type of abuse of the current system where a player can assure that they stay in a position within the ELO rating for a game by playing very few games. This is a kind of 'protect my lead' kind of abuse. In most sports, when a team tries to protect their lead they stop doing what got them the lead with bad results. It's abuse because that player isn't allowing other players to play against them; they're not 'in the mix'. And since that player is now not playing as many games their ELO no longer reflects what their rating should be.

If I understand there are two possible solutions for this.

1) Penalize a player who isn't playing 'enough games' to maintain their true ELO. My thoughts:

-Instead of making a time based penalty, make the penalty based on average number of games per week. Compare each players average against the average of the top 20 players for that game. Any players who aren't at some percentage of that average should lose a game against the mean (1500) player.

So, if the top 20 players averaged 20 games this week, then any player who is not at least, lets say, 20 percent of that number (or 4 games), that player loses to a 1500 player. Then a week goes by and the process repeats. It should only affect that players ELO and not wins and losses or percentage.

2) Reset ELO to 1500 at some interval, like once a quarter. One thing is for sure, that would produce a wild barrage of games during the first few days as players try to be the first to get to the top positions. Probably should stagger each game to reset at different times

[sourisdudesert]

Hi,

Thanks for all your suggestions. There's probably some issue with the current system for sure...

Just to remember the history:

• At the beginning of BGA, there was a simple ELO with no special features.
• Then, we introduce an ELO penalty for some who didn't play for a while (=> back to 1500).
• After few month, it appears that it "broke" the system, because the average of all players was not around 1500 => we removed this system.
• We replaced it by the "do not display in rankings" when did not play since 60 days.
• We also change the way ELO was computed to encourage to play games with the suggested number of players and to give more "elasticity" to ELO during the first games.
• ... so here we are!

Then, to me here's what I've learned:

• The ELO rank is primarily used to determine the level of your opponents. Then, if should reflects the level of your opponents. If your opponents is 2000+, this is a very good player even he didn't play for one year.
• But the ELO rank is also used for games leader boards (#1, #2, ...). Leader boards should reflects the recent performance of a player: if someone is 2000+ but didn't play for one year, there is no reason to make it appear in the leader board.
• As you can see, ELO points are used in 2 different fields, with incompatible goals. The current situation is: we choose to keep ELO rank its original meaning (ie: level of player) and introduce something to improve the leader boards (ie: 60 days delay).
• I really would like to keep this "do not change ELO" idea. BUT, I really think - as you suggest - that we can improve leader boards.

I understand from your post that one of the main criticism of this "60 days delay" is this idea that a good player can maintain a high ranking by playing 1 game every 60 days. This is why I think the temporary "inactivity penalty" is a good idea, if applied on leader board only.

To be continued...

[gdBlaze]

Good points, good think I'm not running things

What about instead of having ELO satisfy two different concepts, have two different ELO's; a player skill ELO and a game rank ELO? The calculations are exactly the same for game results except that the ranking ELO is subject to extra rules to reflect current level of play. Since you wouldn't be affecting the player skill ELO, you could be free to make more interesting rules for ranking ELO without fear of losing the integrity of ELO.