ELO Reset

[Tom_hardy]

I have a proposition to make that came to my mind when evaluating our ELO System in Supremes case. People say its flawed due to several reasons, being the huge amount of games played needed to find your correct ELO as well as the big fluctuation due to luck streaks and being paired with people unfairly cause most havent reached their correct elo yet. That usually takes around 2/3 of the season and people who dont play often never get it. That leads to huge inaccuracy. So my first proposal is definitly:

1. Do not reset the elo after a Season.

I know, people like fresh starts, but lets be honest, its the biggest problem this System has. Just stay where you are and improve if you can. ELO only works with as many games played as possible. No other game with a matchmaking rating does this every few months, maybe every 5 years but in general there is no need for it since it autoadjusts.

2. Define finishing ELO by averaging the ELO of the last x amount of games played.

This way Lucky or unlucky streaks lose a lot of importance and again we get a better estimate. The number shouldnt be lower than 30.

3. You only get ranked this season if you have at least x games played.

I think that rule is necessary to avoid lazy people sitting on their elo once they reached it. If you wanna win a season, you have to participate, simple as that. I would propose 100 games.

People keep saying how flawed our System is and dont take it serious. So lets make it right, once and for all.

[FateStayNight]

I highly support this, allthough it will be annyoing for people in bot 30 i can imagine. But competitionwise its way more honest than the actual system.

[bezdak]

+1

[Diablo_]

I guess the current Elo system isn't perfect (mainly because seasons are a bit too short), but I'm not sure your suggestions would be good.

1) That's too extreme. We had seasons with huge Elo differences between players (mostly few bot and top players) and it would be hard to catch up the next season even if the real skill changes. What I could imagine might be useful is to not reset the Elo completely but to compress it towards 1000 Elo, simple example could be 1000+(elo-1000)/2, more complex examples could use logarithmic functions to further reduce too big differences. This way people would likely start closer to their real ratings while they are still close enough together to allow big changes.

2) Interesting idea but I think this will fail due to technical reasons. Afaik ENT isn't storing any historical Elo data so it would need some possibly rather large changes to make this possible.

3) With current system this is quite irrelevant, with the suggested changes this would be needed, yea.

[Tom_hardy]

@diablo_

Too extreme for what precisely? These huge differences resulted from random teams with 15/15 balance if i remember correctly. Most of the seasons its just a few players who really manage to get far away from the 1000, be it top or bot. If one season is enough to get there, it should also be enough to change if you highly improve all of a sudden, which is unlikely enough as it is.
Using a lighter reset is a possibility, but if we wanna be a competetive league, I honestly dont see the point. Whats wrong with people starting at a point that represents their total prior Performance, be it good or bad?

The implementation of the average games could indeed be problematic, depending on how the system works. Would be nice to get some insight on that.

Edit: I honestly think it can even be more rewarding for the below 1000 ELO players. If you improve your real elo over a season from 700 to 850. You actually can see you improved and gained ELO every step of the way. If you start again from 1000 you will again just gradually see ELO taken away like last season and just end a little sooner.

[BoretkPanda]

i like

but i can also see people quit lihl cause of mad by stucking at 500 elo

[KiwiLeKiller]

but i can also see people quit lihl cause of mad by stucking at 500 elo

Would be great

[Meshtar]

I agree with this proposal, but I'd like a fine-tune addition. People get ranked and rewarded according to the seasonal elo, same as before but they keep their total elo and then maybe implement a yearly award for people with most total elo. So basically every player has 2 elo stacks, 1 that represents his total elo that doesn't get reset every season (maybe reset it yearly or so) and one that represents his seasonal elo, so as Tom said everyone can see how they did current season and how they improved overall or not. People with not enough games per season wont get ranked that season but will still get benefit in their total elo. That way if someone plays 15 games total for a season and just gets lucky enough he doesn't get ranked in top 10 because of that .
The only decision that needs to be done then is if to votebalance by total elo or by seasonal elo (I think total)

[Don_Killuminati]

but i can also see people quit lihl cause of mad by stucking at 500 elo

Would be great

Kiwi finally find his way to quit lihl

I think its a good idea maybe need to be a little bit more adjusted as dia said something like elo+1000 /2 as mmr reset for lol .

[Tom_hardy]

@Meshtar

this would more or less destroy the whole point of it since you are supposed to be matched and valued according to your real ELO by votebalance, not the new, seasonal ELO. So its either you match people according to that, but start from 1000 which would give a weird and shifted view of things, or you match people according to their seasonal ELO which is like now, just that you somehow add the ELO parallel to your previous ELO? That really wont make a lot of sense and will result in people going negative, or 2000+ in ELO overall.

We either have to keep or change the System, this double way wont help imo.

[Diablo_]

@diablo_

Too extreme for what precisely? These huge differences resulted from random teams with 15/15 balance if i remember correctly. Most of the seasons its just a few players who really manage to get far away from the 1000, be it top or bot. If one season is enough to get there, it should also be enough to change if you highly improve all of a sudden, which is unlikely enough as it is.
Using a lighter reset is a possibility, but if we wanna be a competetive league, I honestly dont see the point. Whats wrong with people starting at a point that represents their total prior Performance, be it good or bad?

Too extreme because Elo should at the end of the day represent the current skill. There have been large gaps in many seasons, not only at the 15/15 ones. Just take the current season as example. Ranz is #1 with 1600 Elo, Magic #2 with 1250 Elo. Let's assume Elo precisely represents skill for simplicity. Let's also say the season ends with these stats. Then ranz will have a whooping 350 Elo advantage and even if e.g. ranz becomes a bit less skilled (because he plays less or can't handle new meta as well or w/e) and magic becomses a bit more skilled, ultimately surpassing ranz in skill, he would still have to gain 350 more Elo than ranz, which will be really hard if he is just slightly better than ranz now. If Elo would be compressed to some degree then ranz would still have the advantage but the current skill would have a bigger impact and would allow Elo to faster represent it.

Or also imagine a season where many good players barely play and one guy dominates the whole season ending with 800 Elo advantage. This would basically guarantee him to also win the next season, no matter if he would still be the best player or not.

Compressing Elo does no harm and only makes a new season reflect the current performance stronger than the past performance. Besides that I think it will also increase the fun since it adds fresh air to the league (just like the current stats resets do).

[Meshtar]

@Meshtar

this would more or less destroy the whole point of it since you are supposed to be matched and valued according to your real ELO by votebalance, not the new, seasonal ELO. So its either you match people according to that, but start from 1000 which would give a weird and shifted view of things, or you match people according to their seasonal ELO which is like now, just that you somehow add the ELO parallel to your previous ELO? That really wont make a lot of sense and will result in people going negative, or 2000+ in ELO overall.

We either have to keep or change the System, this double way wont help imo.

If you read my whole post I said votebalance matching should be done according to total elo, the one that doesn't get reseted every season

[Tom_hardy]

@diablo_

First of all, I dont disagree with your idea completely, it could be a nice way to incorporate both, the fresh start without killing all the build up relations.

I still cannot agree fully with your concerns. Lets start with the current season. To me this is exactly what I am talking about. Ranz played more games, ca. 2.5 times the amount of games magic played, thus not representing the real ELO yet. Magic just didnt play enough. Not only him, the whole thing is completely messed up. If you look at the final standings from the last season, you can see it being a lot smoother. Still not being perfect, but we are getting there. Your point that ELO should represent the Real skill and if the skill changes it takes too long to adjust doesnt really hold. Nobody is gonna go from zero to hero in a day, same with the other way around. If the difference is small to begin with, then the Real ELO difference is also gonna be small, once enough games are played. Thats why the current standing with Ranz and Magic are never gonna happen if we keep tracking the ELO.
And if the small skilldifference turns around, so will the ELO.

Im not sure about the likelihood of that case, but fair enough, lets assume somebody found the scroll of truth and gets from 600 real ELO to 1300 real ELO just in terms of skill in a day. You are right, the time is a little longer to reach it, but what are the chances that exactly happens at the end of the season? If it happens in between, he is not gonna finish at 600 anyway, rather being a closer to his actual level.

Your next example is talking about the missing good players for a seaon and the unfairness for the next season. Yes of course, ELO only works relative to your competetors, no doubt about that. Again im having likelyhood issues, but lets assume, one season is gonna be a really weak average, with one person farming all of them, then instantly at day one of the next season the good players are back.

First of all, he is gonna farm People with low average ELO, so the games wont be 15 15 in the entire old season probably.
Now lets go to the next season. He is constantly gonna be paired against superior teams with a higher true ELO for at least 100 games. You dont think Thats gonna adjust his Real ELO to a reasonable degree? Given he cant be terrible since he won that previous seaon so handily.

Again, I dont wanna say your idea is not good as well or should not be done. Maybe its the way to go. I just cant help but discuss your concerns, since they seem all the more reason to me why its a good idea to me.

[Tom_hardy]

@Meshtar

Actually I did. I just answered for both possible scenarios, maybe not clear enough.

Lets assume everybodys total ELO is perfectly reflecting their skill After enough games. Now a new season starts, everybody is at 1000 again, but gets matched according to their total ELO. If nobody changes his personal skill, thats means, everybody is gonna stay exactly at 1000 Seasonal ELO in Average since every game is gonna be completely fair playerwise from the beginning. All you could see in Seasonal ELO would be personal improvement or getting worse. So if somebody with 600 total ELO improves to 700 total ELO over a Season, he will end with 1100 Seasonal ELO. Somebody with 1200 total ELO who did not change his skill would be at 1000, ranked worse than the 700 Total ELO guy. Such a seasonal ELO would be closer to the derivative of skill than the actual skill. I hope this is understandable.

[n1ll3]

I am a fan of Toms idea.

I can agree with all he said so i support this!