🌿ELO System Updates

[Silene]

I don't lose my ability to play a game though just because I haven't played it in 3 months. You basically only play one game here so it would be no problem for you to keep your status. The purpose of the elo system is to rate how strong of a player you are. You believe the real master level for hanabi is 900+. With this current change if a 900 elo hanabi player didnt't play for 1yr their elo would drop to 590. This player has the exact same skill they had prior but now you wouldn't even allow this player to sit at a table with you. I just played a few games I haven't played in months this week. Just as good as ever. I don't want to feel like I have to play this game by this date or lose this amount of elo.

The purpose of elo decay is to remove those at the top who stopped playing. To show the rankings in a way that the top is showing active players an their relative elos. Also, some past systems allowed players to artificially inflate their elos, now they sit at the top, not playing. Got any other ideas how to do it?

Some have already been presented in this thread. To reiterate:
solution 1: Don't show inactive players in the boards. Let them keep their ELO on their profile, but filter them out in the boards.
solution 2: Don't apply it to everyone, but only to those top players (i.e. find top 20 active player and only do decay to inactive players above the lowest of those). In most games ~300 ELO is nowhere near any ranks. That affect hundreds/thousands of players who are not in any way meddling with the leader board.

I'd say that your skill would likely not be the same if you have not played for a year. The game meta might have changed, newer strategies beating what you have learned, so your old skills would not translate to the same elo level as before. Is it 40% less? Probably not. But you should still prove in a few games that your skill is what is was before. As I mentioned in another post before, Glicko type of system would have a decay in a sense that it displays your elo with a degree of uncertainty. But the moment you play and win, it can jump back up to your old elo.

Maybe the system needs tweaking or better explanation, but the system needs changes. Just taking away points does not sound right but decaying elo in a way that displayed value is smaller the less you play, sounds fair.

Glicko would indeed be awesome - then the leaderboard could only show people whose uncertainty value is good enough. But implementing it - that's a very big change. I guess the K-Factor is used to simulate one of the main ideas. Maybe improving K-Factor-usage might be easier to do.

However it has been mentioned that there's decay Glicko, but I think so far it's not been made clear enough that Glicko does NOT decay rating. It ONLY decays the uncertainty (in Glicko System, everyone has those 2 numbers, not just 1 as ELO). And it surely does not assume, that after a year of absence, you're probably weaker. It only assumes that it is more difficult to estimate your skill now. It works in both directions. It also accommodates the possibility that in your year of absence, you became much better (maybe you've been absent because you've been on training camp or whatever).

[T72on1]

I don't get it. One already gets removed from the ELO ranking when not playing a game for a certain period (I think 3 months). So why have ELO decay on top of that?

[Stroom]

Some have already been presented in this thread. To reiterate:
solution 1: Don't show inactive players in the boards. Let them keep their ELO on their profile, but filter them out in the boards.
solution 2: Don't apply it to everyone, but only to those top players (i.e. find top 20 active player and only do decay to inactive players above the lowest of those). In most games ~300 ELO is nowhere near any ranks. That affect hundreds/thousands of players who are not in any way meddling with the leader board.


Glicko would indeed be awesome - then the leaderboard could only show people whose uncertainty value is good enough. But implementing it - that's a very big change. I guess the K-Factor is used to simulate one of the main ideas. Maybe improving K-Factor-usage might be easier to do.

However it has been mentioned that there's decay Glicko, but I think so far it's not been made clear enough that Glicko does NOT decay rating. It ONLY decays the uncertainty. And it surely does not assume, that after a year of absence, you're probably weaker. It only assumes that it is more difficult to estimate your skill now. It works in both directions. It also accommodates the possibility that in your year of absence, you became much better (maybe you've been absent because you've been on training camp or whatever).

This is all just players caring about their elo score and comparing it to others rather than actually caring about that showing their relative skill to others. Old player elo vs new elo is not comparable. So many factors have been in the system that it doesn't make sense.

With glicko, you display player "RANKING" = "TRUE_ELO" - "UNCERTAINTY" - so from the point of view of the players, you only see the calculated value. No other player will see their true elo anymore. And that is fine. No-one should care about what their true elo is anyway. What actually matters both for setting up the game and any further information, is the calculated value.

If you have not played for a long time, come back and start playing, your true elo will rise or decline based on winning or losing and as the uncertainty decreases, the displayable calculated value will also naturally become what it is supposed to be. It's a nice system. The current decay seems to want to do the same thing, maybe a bit too drastically. Or they didn't really explain very well how it would work.

As I said before, as the system has changed over time, an elo value from 4 years ago might not be at the same level as it is now. Maybe in the past, the system somehow favored some style of playing that is not available anymore so players gained such high elos that are not reachable anymore... so the elo value has no real meaning today anymore. Hanabi abandoning players for instance. No way someone can get 1600 elo today.

[Silene]

With glicko, you display player "RANKING" = "TRUE_ELO" - "UNCERTAINTY" - so from the point of view of the players, you only see the calculated value. No other player will see their true elo anymore. And that is fine. No-one should care about what their true elo is anyway. What actually matters both for setting up the game and any further information, is the calculated value.

If a platform decides to show that difference rather than true_elo, that's fine. But the essential thing for the system to work is, that the true_elo still exists in the background and will be used to process new results. The true_elo only changes depending on results and never from absence. So yes, your shown value is decreasing over time (since it's true_value MINUS uncertainty). This will still lead to a rapid pull up, should you come back and win while at the same time protecting your opponent from losing too many points to you (because your still very high true_elo is used to calculate their new rating and since your ranking is so uncertain, their loss will even be weighted very low).

Just reducing the score without having these 2 essential features is just a combination of the worst from both worlds and NOT a step in the right direction!

[ChiefPointThief]

The purpose of elo decay is to remove those at the top who stopped playing.

The purpose of me stating that I am ranked like 300th in that game was to hint at the fact that I am nowhere near the top (top currently 944). In fact there are roughly 2500 players in stone age that would be affected by this change. I agree w/ a lot of what silene has posted. Who cares if someone has 400 elo in stone age and hasn't played in 6 months. Who is that bothering? Why target someone who is 1500th on the leaderboard list if the purpose of this change is to for players at the top.

Another problem with this is the filter option. I have seen plenty of tables that say strong players and below. These ppl clearly do not want to play with ppl that are too skilled for them. With this change you will have a bunch of experts demoted to strong. Furthermore when the reduced expert defeats the other players as they should because they are truly an expert they will be taking more elo from these players than they should.

[Mr_Magic]

Perhaps BGA should implement true Glicko behind the scenes and then swap over to it after an arena season or two once everyone has got a proper new glicko rating . The uncertainty variable in glicko is an awesome way to solve most of this problem

[h_illes]

Another issue with the power levelling fix is the following situation. Take two strong players who play against each other regularly (for whatever reason, small player field etc.). They both win close to 50% of the games between them, but sometimes one of them gets a bit of a win streak going, sometimes the other. With the current fix, during a streak, the loser gets full points reduced, while the winner gets less than full points added. So in the long run, both of their ratings go down - just from playing each other. This is a serious *, and a strong motivation against playing. Please don't do this. The current power levelling fix is a bad mistake which should be fixed as soon as possible. Go with the K factor reduced over a win streak instead, that achieves a similar effect without the harm.

*Moderator edit: please refrain from swearing on BGA, this is a family-friendly platform.

[ChiefPointThief]

The purpose of elo decay is to remove those at the top who stopped playing.

The purpose of me stating that I am ranked like 300th in that game was to hint at the fact that I am nowhere near the top (top currently 944). In fact there are roughly 2500 players in stone age that would be affected by this change. I agree w/ a lot of what silene has posted. Who cares if someone has 400 elo in stone age and hasn't played in 6 months. Who is that bothering? Why target someone who is 1500th on the leaderboard list if the purpose of this change is for players at the top.

Another problem with this is the filter option. I have seen plenty of tables that say strong players and below. These ppl clearly do not want to play with ppl that are too skilled for them. With this change you will have a bunch of experts demoted to strong. Furthermore when the reduced expert defeats the other players as they should because they are truly an expert they will be taking more elo from these players than they should.

[Stroom]

The purpose of elo decay is to remove those at the top who stopped playing.

The purpose of me stating that I am ranked like 300th in that game was to hint at the fact that I am nowhere near the top (top currently 944). In fact there are roughly 2500 players in stone age that would be affected by this change. I agree w/ a lot of what silene has posted. Who cares if someone has 400 elo in stone age and hasn't played in 6 months. Who is that bothering? Why target someone who is 1500th on the leaderboard list if the purpose of this change is to for players at the top.

Another problem with this is the filter option. I have seen plenty of tables that say strong players and below. These ppl clearly do not want to play with ppl that are too skilled for them. With this change you will have a bunch of experts demoted to strong. Furthermore when the reduced expert defeats the other players as they should because they are truly an expert they will be taking more elo from these players than they should.

Why does your elo even matter to you? This is a pointless number that you can get back any time just by playing. Also, elo gained in the past is not necessarily equivalent to elo gained today. There are more players, the meta can change etc.

A general rule that applies for everyone would be easier to implement. If the cutoff is at top 10 and someone is in the top 10 loses 10%, the 11th player would suddenly be MUCH higher than him, probably even becoming top 1. Would that be fair?

Also, nothing would stop you from creating a new player with 0 elo and start playing with weaker players. A string player occasionally entering games with weaker players isn't probably that big of a deal as after 1-2 games, the ranking would go up again.

What really is needed, is a good system. Not players completely rejecting the solution without providing their own alternatives. I proposed Glicko and seems like people approve of it.

[Ranior]

The impacts of the power leveling "fix" are horrible and it really does need to be undone as soon as possible.

Perhaps there is some other option that can fix this supposed problem, but as is it's doing really bad things.

As others and myself have pointed out here are some of the issues:

--For smaller player pool games, and high player count games, you can wind up playing a similar opponent frequently. If you get on a win streak against someone, it heavily incentivices you not to play against them if you can. But for things like arena you cannot avoid them, and for some smaller player pool games, it may mean you can't get to play a game at all in real time if you're trying to avoid them. Or if you're playing turn based you have to try to exclude them from your tables?

--It's very hard to tell exactly how many games you need to be separated to avoid the decay for consecutive wins, but it makes reaching high elo very hard to impossible in smaller player pool games. There are only so many players playing live 4p Saint Petersburg, if I get reduced elo from consecutively beating players, but full losses, it's basically impossible to compete on a fair level with those who already have high elos and just don't play many games. (Or it's just harder to compete against 2p only players who have an easier time avoiding repeating opponents within X games)

As with too many things on this site related to competitive play, you've really botched it once again. The early arena seasons were plagued with silly issues (abusing turn based table finishes, losing elo due to ties in games with quitters, etc). This roll out of adjustments is also plagued with issues that are easily spotted by your player base. Perhaps you should ask for some more feedback from trusted and experienced users before doing some of these things.

Finally, it does also seem that these elo changes have reintroduced old issues that were corrected. Which again is annoyingly persistent issue for this site. I adore BGA, but I am baffled sometimes by how the admins operate on some changes. Anyhow that issue is multiplayer tables where there is a quitter--the game considers this a tie among all players, which can cause high rated players to lose elo due to the tie. This was corrected in the past but seems to have resurfaced with these changes--you need to make it so that players cannot lose elo when someone else quits the game.