The new ELO-based ranking system

gamerman01

@mr_stucifer said in Proposal for a new, ELO-based, ranking system:

Not to call anyone out, but entering over 500 games I see people playing the bottom-ELO players and half the time that player gets a below-average bid, and they are always losing those games.

(The following is, I’m launching from your point, it is not at all a retort back to your post.)

Thank you, I was going to make this point yet again, but now you’ve given me a head start.

To the complaint that you can lose points with a win.
Those games shouldn’t even be played, because they are not competitive.
To the complaint that you can’t learn unless you play somebody way better than you, I tend to disagree. Watch someone else’s game, or play someone a tier higher than you. A bug doesn’t learn much getting hit by a windshield.
A stronger argument from me is that the better player is putting a lot of time into a game that, as you said, is always being lost, instead of playing a much more competitive game.

The thought behind the complaint of losing points for winning games is that everyone should be able to play everyone. I’m just saying it’s not that simple. In my opinion it is not a great improvement to give the #1 +1 points for beating the bottom player. What a waste of time. If you don’t want to lose points and the #1 player wants to play the bottom player, you could always play that game outside the league.

I’m fine with the system always giving some increase for winning, how can you argue against that?
(Except for the argument that those games are pretty much a waste of time. This isn’t a 20 minute game of chess)

New system, ELO adjusting based on current ratings is great. I’m just saying that an upper player is going to pound a low player every. single. time. And that with ELO, appropriately, top player will get about 1 point, which is appropriate, and my point is the #1 player still won’t play the bottom player if he’s wanting a higher rating, the time sink is totally not worth it, so there will always be the same complaint - top won’t play the bottom. It’s because of the time commitment.

But it’s not about the points. A top player might have a blast destroying/teaching a much lower player. I guess I’m not saying those games can’t and won’t happen, I’m saying the rating system shouldn’t reward such games (and neither the past or the future one do)

Stucifer

@gamerman01 said in Proposal for a new, ELO-based, ranking system:

The thought behind the complaint of losing points for winning games is that everyone should be able to play everyone. I’m just saying it’s not that simple. In my opinion it is not a great improvement to give the #1 +1 points for beating the bottom player. What a waste of time. If you don’t want to lose points and the #1 player wants to play the bottom player, you could always play that game outside the league.

This is an excellent point and one I hadn’t given enough thought to prior to you pointing it out, thank you :)

farmboy

I think PPG worked quite well for our purposes. It produced standings that we could use yearly for the playoffs and it accommodated players playing as few as 6 games (or less) and (in a few rare cases) more than 30. PPG does over time reduce the impact of new games, but with only the occasional exception, no one played so many games in a year that winning a couple of games against strong opponents wouldn’t be meaningful for your final ranking. The final standings were fairly consistent with how people played. Although I know I don’t fully understand some of the concerns raised, it did seem to me that there was a misunderstanding around how rankings were determined. No one’s ranking was ever tied to one game. It was either tier 3 (if they were new and had played less than 3 games), their previous year’s ranking (if not new), or, once they had played at least 3 games, their average PPG.

I’m being won over to ELO, not because I think it will do a better job of those things in a given year, but because I now like the idea of lifetime rankings. And I’m enjoying seeing all this data that is coming with the transition.

farmboy

For playoffs, we can try it and see how it goes. I continue to have concerns though. I understand that if someone new wins 8 games in a row their ranking may jump into the top 8 or if someone normally strong loses a bunch of games, they will drop down. But I still think given the small number of games played in a year we may find that things get more locked in. Players that don’t quite do that well might still have done well enough in a year to join the playoffs (but may still end up on the outside). And I don’t expect top players to ever do that badly.

And I expect that as we go back in time, we are going to see a few players ELO moved into the low-mid 2000s. I understand that ELO growth slows as players move up, but between 2018-2020, JDOW wins 34 and loses 1, Adam wins 55 and loses 5, and AD wins 88 and loses 24. Maybe I don’t fully appreciate how an ELO will accommodate that data (and minimize its impact), but I suspect it will make catching those players close to unreachable in the short term.

We also might find with the numbers that are interested, and the number of players that move on each year, that newer players aren’t trapped on the outside so we may not need to over think it.
But one option, if there are issues, is to use ELO to determine one’s ranking (and one’s tier), but PPG in a year is still used to determine playoffs (where the points per game would be tied to the tier produced by the ELO ranking). If the sheet is automated, maybe that doesn’t complicate it too much.

Stucifer

but I suspect it will make catching those players close to unreachable in the short term.

As long as they get a game with them the ELO recalibration will be quick. If a player with 0-3 games completed beat one of them with a 600 rating difference that would result in about -65 points for them and +130 for the new player. Do this three times and the points reduce somewhat but they should still be around an 1800 ELO after just 3 games, clearly a top-tier player with a rating similar to GeneralDisarray, Ghostglider, ksmckay, and Pejon_88

farmboy

@mr_stucifer I agree that if someone plays the top players and has a winning streak, I’m sure they can get in in just a few games. But if someone plays at the level of the top players then they might win 4 games and lose 4 games. In the current ppg system they would likely be competitive enough to make the playoffs. But in an elo I suspect that very good players would need more games because they are going to also lose some games.

gamerman01

@Adam514 said in Proposal for a new, ELO-based, ranking system:

I don’t think rating changes should be dependent on bid. Both players consented to that bid, presumably for balance reasons. It’s what both players are satisfied playing with. No need to put a rating factor on that.

Of course I totally agree with this about ratings and bids.

I’d be interested in getting a higher rating for how fast my opponents give up, but that’s a bit problematic as well.

🤓🤐💪

MrRoboto

Some concerns about the ELO system, particularly for playoff ranking, are absolutely understandable.

And I absolutely have to admit: that particular case that farmboy created (a new, unknown player going 4-4 against the top players) would be slightly better represented by the old/current PPG system!
He would collect 4x4 + 4x8 = 48 points = 6.0 PPG and would put him in Tier M, on rank 7

And unfortunately, in ELO he would probably be rated a bit too low for achieving 4-4 against the best players:!

However, as someone (I think it was even farmboy itself) already pointed out:

There is ALWAYS a case where a system doesn’t work perfectly, you can always create special circumstances.
For example in the PPG system, Karl7 is #2, which is certainly questionable since all of his wins are against mediocre or low-skilled players (myself included!)
I do think that this particular case is very theoretical and not very likely to happen (but you can correct me if I’m wrong and it happened before!).
That player would still be #4 (of the active players) and rightly make the playoff. Even if Gamerman, Pejon and Bombsaway all manage to complete 6 games, that player would still be included
I admit that the system works best for players who complete around 15 games or more so with more games coming in for that newcomer, the system won’t fail him over time

Concerning bids impacting the rating or not:
You convinced me! Enthusiastic and energetic me thought this was a cool idea and I wanted the system as sophisticated as can be. But I agree: it’s overcomplicating things and bids are (at least so far) not used to balance things, but to agree on sides.
Scratch that idea!

MrRoboto

One more thing:

I designed the specific math and factors (K-factor and F-Factor) with the results I have seen and the experience I had here.
What I’m saying: This particular ELO system is not a simple 1:1 copy of chess, in fact I got inspired by the old League-of-Legends system (LoL is the most popular esport-game).

So it is designed on real, past results. But if we notice that it doesn’t serve our particular purposes or if we see some players ranked unfairly, we can always tweak the math behind it to better represent our community!

gamerman01

It’s hard to get it right with a new player.

Dawg just defeated donutgold and with ELO, got a lot of points. No matter what donutgold goes on to do, Dawg has the points (for lack of a better way of putting it)

So Top players could prey on new players because 1500 is probably higher than they would normally be (in all my years of experience), and bottom players could prey on them as well.

Actually, all players could benefit from getting to the new player, if new players are below average skill/experience.

It’s an issue in “my” system as well, just handles it differently.

Maybe it’s what we want - everyone wanting to break in the newbie.

gamerman01

@MrRoboto Agreed - great post

oysteilo

I think maybe it is apropriate to do a recap of where we are right now. Are we proposing to start everyone at 1500 on jan 1 or are we using past history?

How are we handling play offs and different versions?

In the spread sheets i see unknown players with few games and fairly hig rankings. It is not really a concern, but maybe you should stay at 1500 rating until you have 3 games? Is that possible to include?

gamerman01

Right now we’re definitely on track to roll out 1/1
Everyone starts at 1500 for their first league game, so what you see is going back to 2020. We’re going to go back farther.

Continuing ->

gamerman01

In other words, we don’t start at 1500 each year.

Discussion has started about how to look at year by year, version by version like we’ve had the last 3-4 years. From what I’ve read, it’s already shaping up. We’ll have it well before 1/1

The unknown players (to YOU!, not me!) you’re seeing are probably inactives who played a few years ago. They are irrelevant for 2023 and 2024.

gamerman01

So results are being entered to get data in and see how things look. These are lifetime results, and we’re only back to 2020 so far.
The capabilities are there to determine who was best during 2024 and make playoff matchups, by version. We’ll have it before 12/31. Probably involves some combination of looking at lifelong history and current year results. It’ll be awesome, we’re only beginning to discuss this important topic though.

gamerman01

@oysteilo said in Proposal for a new, ELO-based, ranking system:

In the spread sheets i see unknown players with few games and fairly hig rankings.

Gray bar means it’s been over a year since their last game result. (Another sweet feature from MrRoboto)

MrRoboto

@gamerman01 said in Proposal for a new, ELO-based, ranking system:

It’s hard to get it right with a new player.

Dawg just defeated donutgold and with ELO, got a lot of points. No matter what donutgold goes on to do, Dawg has the points

Right now, games against complete newcomers might be over- or underrated.
These are just momentary snapshots though. Yes, dawg received a lot of points for winning against a player who was probably overrated at 1500. But it won’t take long before he is back to his old rating.

Same would happen if you lose against a 1500 newcomer, who is actually a top gamer in disguise. At that exact moment, you’ll unfairly lose too many points, but the system will bring you back to where you belong reasonably fast.

HOWEVER:
I realize this is an issue on some people’s mind. And there is a simple solution for that. As I said, we can always tweak the math to serve our needs. I just didn’t implement a failsafe against this, because personally I didn’t deem it necessary - but I might be wrong and gamerman has vastly more experience with this community and he seems to think this could be an issue.

I can easily tweak the formula so that games against newcomers give only 50% of the usual points. Or 30%. Or whatever.
What value do you think makes sense?
For how long should a newcomer have that “newcomer” status for opponents?

gamerman01

@MrRoboto said in Proposal for a new, ELO-based, ranking system:

Right now, games against complete newcomers might be over- or underrated.
These are just momentary snapshots though. Yes, dawg received a lot of points for winning against a player who was probably overrated at 1500. But it won’t take long before he is back to his old rating.

Ah, right, very true for someone who keeps on playing

Same would happen if you lose against a 1500 newcomer, who is actually a top gamer in disguise. At that exact moment, you’ll unfairly lose too many points, but the system will bring you back to where you belong reasonably fast.

Yes, assuming I keep on playing

HOWEVER:
I realize this is an issue on some people’s mind. And there is a simple solution for that. As I said, we can always tweak the math to serve our needs. I just didn’t implement a failsafe against this, because personally I didn’t deem it necessary - but I might be wrong and gamerman has vastly more experience with this community and he seems to think this could be an issue.

It would be if someone is trying to maximize their points and looks for opponents with 0 to just a couple games done, but I’m not too worried about that.

The (somewhat minor) concern is more for entering playoffs. If 6 remains the minimum for qualifying (BM), then conceivably someone could inflate their score by playing several or all of their games against the unknowns (newcomers).

I can easily tweak the formula so that games against newcomers give only 50% of the usual points. Or 30%. Or whatever.
What value do you think makes sense?
For how long should a newcomer have that “newcomer” status for opponents?

Shoot… probably no adjustment after thinking a few seconds. Because none of us want to discourage someone getting going in the league. If they’re worth full credit as a win over a 1500, that’s OK because there actually is a little incentive to feast on their 1500. This factor of making it easier and not harder for a newcomer to get games overpowers the concerns above.

No change!

Thanks!

MrRoboto

And some clarification on the ranking itself.
Here is the current one for Balanced Mod:

You can see the legend on the right hand side. This should explain every colour, besides white.
White means a player is active but has not yet completed the necessary amount of games to qualify for the yearly playoffs.
Maybe I’ll add an explanation for white or just colour code it differently. It’s important to me that everyone should understand it without needing an explanation. So I might have to improve the UI here.

The rank in the very first column is relevant for the playoffs!

So if you check out the sheet you’ll see that Sovietishcat occupies the last (8th) spot, with @elche missing it at #9.
However, @Sovietishcat will drop out when some of the white lines above him complete 6 games this year, which seems very likely: @Pejon_88 , @GeneralDisarray , @Booper and @BombsAway all have 5 already (although they could also drop below Sovietishcat)

MrRoboto

@gamerman01 said in Proposal for a new, ELO-based, ranking system:

If 6 remains the minimum for qualifying (BM), then conceivably someone could inflate their score by playing several or all of their games against the unknowns (newcomers).

True, but those players would also take a risk: After all, the newcomer could be a secret god in disguise and then they would lose a lot of points against a 1500.

It’s far from a sure way to game the system and I also doubt that we have players who choose their opponents this strategically to maximize their ranking

The new ELO-based ranking system

Featured Topics

T-shirts, Hats, and More

Suggested Topics

Wiz (X) vs Zar (L+21) BM4 G1 L25

L25 PtV Ghostglider (Axis+15) vs Adam514

PTV 25 Nikola v 666 (axis +12)

L25 PTV Stucifer (X+12) v Nikola1975 (L)

L25 PTV ArtofWar (X+10) v Stucifer (L)

pacifiersboard (X) -vs- fasthard (L +22) BM #2

2024 Playoff OOB Round 1 Pacifiers Board (+42) v Gamerman01

L24 OOB Stucifer (X) v Crockett36 (L+120)

69

17.8k

40.4k

1.8m