AOE-II DE "Updates to Ranked Team Game Elo Calculation"

[JonnyJellyfish]

Other than that, it's not only possible to game the system a little at the very top. Smurfing does occur very frequently, in my experience, on all levels on the ladder. While this will never fully eliminate smurfs, of course, it does make it harder to smurf unless all players on the smurf team decide to tank games, instead of just one. If this does indeed help combat smurfing to an extend, it does give a much smoother experience for casual players. Getting stomped by a team smurfing on the ladder is one of the more frustrating things in my opinion, and it wouldn't surprise me if more people get fatigued from this than from the scenario you described.
Basically, a more fair ladder overall makes for more balanced games, leading to a better experience for everyone.

So I don't disagree with your first point, if you always play with the same teammate then it will be the best balance, unless one of you wants to play with another friend, or just queue up alone. But what if he is on holiday and you want your hit of TG enjoyment - arguably if you queue up knowing that you are underrated, you have become the smurf!

In terms of smurfs I think I disagree. Imo it actually makes it easier to drop points - if a smurf wants to drop a 2k Elo account down to 1200, now he need to do is queue up with a friend on a clean/smurf account at 1k Elo, then resign after 5mins and instead of losing just 16 Elo (assuming a balanced match) I lose 30 (according to the aoe official example). This makes it much easier to drop Elo on a specific account and get to your desired range. The other option in terms of smurfing is just to make a lot of clean accounts and keep queuing up at a "starter" Elo, which is just as annoying, and there is no real control for.

I think that the only way to stop the smurfing is to ban the accounts reported for it. It's not difficult to see if someone queues up with the intent of dropping Elo and that would solve both the 4k Elo people after their ego hits and the smurfs who want to beat people worse than them on a lower level.

 

[dodageka]

They just removed the cap with the latest change, now they are trading way more points than before, soon we will see 2700 players, enjoy.

There never was a cap in the 1v1 ratings, it always was a lower k-factor for higher elos (I liked the lower k-factor btw.). I don’t get what your problem with higher elos is though, as an individual value it’s completely meaningless anyways and for the relative value it doesn’t matter whether the top player has 2600, 2700 or 5500.

 

[HongeyKong]

Yea seems like he and the other posts OP dont understand that elo is only useful when theres comparison, the absolute value is just meaningless. And thats the reason why I made these two post moths ago.
https://www.aoezone.net/threads/how-can-devs-make-aoe2de-more-braggable-to-casuals-outsiders.164753/
https://www.aoezone.net/threads/wha...ost-no-extra-info-than-the-elo-rating.178228/

 

[SouFire]

There never was a cap in the 1v1 ratings, it always was a lower k-factor for higher elos (I liked the lower k-factor btw.). I don’t get what your problem with higher elos is though, as an individual value it’s completely meaningless anyways and for the relative value it doesn’t matter whether the top player has 2600, 2700 or 5500.

The elo system never used a cap or k factor or whatever name you wanna put on it, elo system was copied from chess and there if a 2700 faces another 2700 he gets 16 points, that is the system we had before DE, on DE they added a CAP for 1x1 ladder going from 2200 and increasing with higher ranks, that caused deflation and a bunch of 2500 players that couldn't go higher cause they were trading 1-2 points per win vs similar elos, that was a flaw on DE system, while on TG that CAP or k factor ass you claim was non existent so the sky was the only limit, such cap would have made sense for tg and not for 1x1, so the 5k elo player and even the 3k elo players would have never existed, so don't pretend that things were fine.

But they now removed the cap or you beloved k factor, watch villese he just broke in few matches hera's pb and he is getting even higher!!!

 

[Potkeny]

The elo system never used a cap or k factor or whatever name you wanna put on it, elo system was copied from chess and there if a 2700 faces another 2700 he gets 16 points

According to my quick and dirty searching, the FIDE rulebook has this in it:

K is the development coefficient.
K = 40 for a player new to the rating list until he has completed events with at least 30 games.
K = 20 as long as a player's rating remains under 2400.
K = 10 once a player's published rating has reached 2400 and remains at that level subsequently, even if the rating drops below 2400.
K = 40 for all players until their 18th birthday, as long as their rating remains under 2300.
If the number of games (n) for a player on any list for a rating period multiplied by K (as defined above) exceeds 700, then K shall be the largest whole number such that K x n does not exceed 700.

So it seems DE might really just copied the chess Elo system, with the "cap" being at 2400.

 

[AttilaTheHun]

Nice, we got a new calculation and it looks like the devs did a terrible job. The elo calculation isnt a zero sum game anymore. As result the sytem will become inflated again in no time. It is probably not as bad as the old old system, which was also inflated. But it wont be good for the long run. The devs should go back to the drawing board. This change is no fix at all.

 

[Zbyszek]

The elo calculation isnt a zero sum game anymore.

it isnt?

 

[dodageka]

The elo system never used a cap or k factor or whatever name you wanna put on it, elo system was copied from chess and there if a 2700 faces another 2700 he gets 16 points, that is the system we had before DE, on DE they added a CAP for 1x1 ladder going from 2200 and increasing with higher ranks, that caused deflation and a bunch of 2500 players that couldn't go higher cause they were trading 1-2 points per win vs similar elos, that was a flaw on DE system, while on TG that CAP or k factor ass you claim was non existent so the sky was the only limit, such cap would have made sense for tg and not for 1x1, so the 5k elo player and even the 3k elo players would have never existed, so don't pretend that things were fine.

But they now removed the cap or you beloved k factor, watch villese he just broke in few matches hera's pb and he is getting even higher!!!

As Potkeny already pointed out, the elo-system used in chess uses a variety of k-factors (the k-factor is what determines the amount of points a players get. It is often lowered for more established players to create more stability in the rankings and remove swings. It however does not impact the total elo values in the long run, just how fast people are getting there. The total elo value (in a 1v1 System) of the best players only depends on the relative strength of that player compared to the average elo (which is largely driven by the starting elo, especially because one can create accounts so easily)

 

[dodageka]

That will always remain a problem as long as we dont have separate solo que / pre-made TG ratings, and I wonder if we ever will.


EDIT: Actually even if we have that the problem will still exist unless theres a rating for every single team combination, as people can stack with different people.

Some games do that, and I honestly don’t think it would be a bad idea

 

[AttilaTheHun]

it isnt?

No, it is again a non zero sum game. In the old bugged calculation it also wasnt a non zero sum game. Then they fixed that and it was a zero sum game. But how they changed it some days ago it isnt a zero sum game anymore.

Just do the calculations for a 2500+500+500+500 team vs a 1000 + 1000 + 1000 +1000 team.
If team 1 wins, then the 2500 player almost got nothing since 2500 v 1000 is an easy win for the 2500 player. A 500 winning against a 1000 has a really small change, so they all get almost the full 32 points each. So that side is getting about 90 points. On the other side it is 1000 v 1000 for all 4 players. That is a 50-50, so they all losse 16 points, which is 64 points in total. So this game would generate about +30 elo and it clearly not a zero sum game.

 

[KolyaKrasotkin]

Just do the calculations for a 2500+500+500+500 team vs a 1000 + 1000 + 1000 +1000 team.
If team 1 wins, then the 2500 player almost got nothing since 2500 v 1000 is an easy win for the 2500 player. A 500 winning against a 1000 has a really small change, so they all get almost the full 32 points each. So that side is getting about 90 points. On the other side it is 1000 v 1000 for all 4 players. That is a 50-50, so they all losse 16 points, which is 64 points in total. So this game would generate about +30 elo and it clearly not a zero sum game.

Doesn't it work like as follows (?):
Based on team average rating it is decided how many points are traded. In your example it is 1000 vs. 1000, so 16 points times 4 players equals 64 points.
And then, based on individual elo within the team, these points are distributed (gain and loss) among the players?
That would be a zero sum game.
But I am not sure, whether it's implemented like that.

 

[AttilaTheHun]

Doesn't it work like as follows (?):
Based on team average rating it is decided how many points are traded. In your example it is 1000 vs. 1000, so 16 points times 4 players equals 64 points.
And then, based on individual elo within the team, these points are distributed (gain and loss) among the players?
That would be a zero sum game.
But I am not sure, whether it's implemented like that.

They didnt explained something like this, so i dont think they are using something like this. Based on the post it is just your elo vs the average elo of the enemy team. Nothing else.

We plan to correct this problem by using a player’s individual team game Elo, instead of their team average, and calculating that against the average Elo of the enemy team.

So my guess would be they dont use something like this to make it a zero sum game, which is clearly an oversight from the devs.

Your suggestion is already an easy fix against the inflation, and easy fix to make it a zero sum game again.

 

[knutan96]

They didnt explained something like this, so i dont think they are using something like this. Based on the post it is just your elo vs the average elo of the enemy team. Nothing else.



So my guess would be they dont use something like this to make it a zero sum game, which is clearly an oversight from the devs.

Your suggestion is already an easy fix against the inflation, and easy fix to make it a zero sum game again.

• The example given on the post shows it being zero sum https://www.ageofempires.com/news/updates-to-ranked-team-game-elo-calculation/
• Team 1:
  • Player A at 2000 Elo
  • Player B at 1000 Elo
• Team 2:
  • Player C at 1500 Elo
  • Player D at 1500 Elo
• If Team 1 wins:
  • Player A on Team 1 will gain 2 Elo (as if a 2000 Elo player beat a 1500 Elo opponent)
  • Player B on Team 1 will gain 30 Elo (as if a 1000 Elo player beat a 1500 Elo opponent)
  • Player C on Team 2 will lose 16 Elo (as if a 1500 Elo player beat a 1500 Elo opponent)
  • Player D on Team 2 will lose 16 Elo (since their rating is the same their teammate, they lose the same number of points in this scenario)

 

[AttilaTheHun]

The example given on the post shows it being zero sum

That is because it is one of the few scenarios in which the outcome is zero sum:

Team A:
X+Y
X-Y

Team B
X+Z
X-Z

If these are the teams for a given X, Y, Z, then the outcome is zero sum. The example of the devs fits this, thus the outcome is zero sum. In almost all other cases it isnt a zero sum.

The devs did just show this example this way and didn't mentioned anything about rescaling the outcomes to make it zero sum.

 

[ninja14]

The system which voobly has works perfectly so why don't they just copy what works.

 

[RAY]

The system which voobly has works perfectly so why don't they just copy what works.

That's what they did

 

[ninja14]

That's what they did

But voobly system is indeed zero sum for a single match, while people are saying that’s not the case here

 

[AttilaTheHun]

That's what they did

No, they didn't.

 

[T-West]

I random-queued into a team game for science and checked how the Elo changed.

Winning Team Average: 1417.75 -> 1433.75 (+16)
- T-West 1558 -> 1568 (+10)
- LALO 1376 -> 1394 (+18)
- Petard Pan 1378 -> 1396 (+18)
- Optimus 1359 -> 1377 (+18)

Losing Team Average: 1411 -> 1395.5 (-15.5)
- CORP_Robledo Frias 1355 -> 1342 (-13)
- CORP_Qeetsa 1362 -> 1349 (-13)
- CORP_Anakin 1436 -> 1419 (-17)
- CORP_NicoMonte31 1491 -> 1472 (-19)

So it is possible for different teams to gain/lose different total Elo values (i.e. individual team games are not "zero sum").

However, the "Elo inflation" problem from DE's original Elo system was that there always was a "net gain" of Elo, regardless of which team won and which lost. In the current system it should be possible also to have a "net loss" of Elo, where the losing team members lose, in total, more Elo than the winning team gains. The update post didn't explain this clearly since its example was "zero sum".

(If anyone has stats for how the Elo changes on Voobly and HD, that would be good to check.)

 

[nimanoe]

Losing Team Average: 1403.5 -> 1395.5 (-8)

Should be 1411 -> 1395.5 (-15.5)

 

[T-West]

Should be 1411 -> 1395.5 (-15.5)

fixed, thanks
Should add that in total, the winning team gained 10 + 18 + 18 + 18 = 64 points. While the losing team lost 13 + 13 + 17 + 19 = 62 points.
Although it may be possible that Elos are decimal numbers that only are displayed as ints. In which case there may be some rounding involved, with both teams actually gaining/losing the same amount.

 

[HongeyKong]

I random-queued into a team game for science and checked how the Elo changed.

Winning Team Average: 1417.75 -> 1433.75 (+16)
- T-West 1558 -> 1568 (+10)
- LALO 1376 -> 1394 (+18)
- Petard Pan 1378 -> 1396 (+18)
- Optimus 1359 -> 1377 (+18)

Losing Team Average: 1411 -> 1395.5 (-15.5)
- CORP_Robledo Frias 1355 -> 1342 (-13)
- CORP_Qeetsa 1362 -> 1349 (-13)
- CORP_Anakin 1436 -> 1419 (-17)
- CORP_NicoMonte31 1491 -> 1472 (-19)

So it is possible for different teams to gain/lose different total Elo values (i.e. individual team games are not "zero sum").

However, the "Elo inflation" problem from DE's original Elo system was that there always was a "net gain" of Elo, regardless of which team won and which lost. In the current system it should be possible also to have a "net loss" of Elo, where the losing team members lose, in total, more Elo than the winning team gains. The update post didn't explain this clearly since its example was "zero sum".

(If anyone has stats for how the Elo changes on Voobly and HD, that would be good to check.)

Actually, AttilaTheHun is right, if they are using the elo formula, each games expected value of elo change will be positive.

The sum of elo change of both outcomes will be zero. However, as the likeliness of the two outcomes arent 50/50, there will be net gain.

 

[HongeyKong]

Here are two examples

Example 1: 1800 1200 vs 1500 1500
If Team 1 wins
A gains: 1-1/(1+10^(((1500+1500)/2-1800)/400)) K = 0.1510 K
B gains: 1-1/(1+10^(((1500+1500)/2-1200)/400)) K = 0.8490 K
C loses: 1/(1+10^(((1800+1200)/2-1500)/400)) K = 0.5 K
D loses: 1/(1+10^(((1800+1200)/2-1500)/400)) K = 0.5 K
=>Net = 0

If Team 2 wins
A loses: 1/(1+10^(((1500+1500)/2-1800)/400)) K = 0.8490 K
B loses: 1/(1+10^(((1500+1500)/2-1200)/400)) K = 0.1510 K
C gains: 1-1/(1+10^(((1800+1200)/2-1500)/400)) K = 0.5 K
D gains: 1-1/(1+10^(((1800+1200)/2-1500)/400)) K = 0.5 K
=> Net = 0

==>> Expected net change = 0





Example 2: 2200 1200 vs 1500 1500
If Team 1 wins
A gains: 1-1/(1+10^(((1500+1500)/2-2200)/400)) K = 0.0175 K
B gains: 1-1/(1+10^(((1500+1500)/2-1200)/400)) K = 0.8490 K
C loses: 1/(1+10^(((2200+1200)/2-1500)/400)) K = 0.2402 K
D loses: 1/(1+10^(((2200+1200)/2-1500)/400)) K = 0.2402 K
=>Net = +0.3860 K

If Team 2 wins
A loses: 1/(1+10^(((1500+1500)/2-2200)/400)) K = 0.9825 K
B loses: 1/(1+10^(((1500+1500)/2-1200)/400)) K = 0.1510 K
C gains: 1-1/(1+10^(((2200+1200)/2-1500)/400)) K = 0.7597 K
D gains: 1-1/(1+10^(((2200+1200)/2-1500)/400)) K = 0.7597 K
=> Net = +0.3860 K

==>> Expected net change = +0.386 K



EDIT: Fixed some mistakes


EDIT 2: Here are the numbers for the OLD OLD algorithm
Example 1: Expected net change +0.698 K
Example 2: Expected net change +0.831 K

So as bad as this new algorithm could be, it should still be better than the old old algorithm.

 

[ninja14]

Why can't they make the ELO gain for one team same as ELO loss for other team.

You can distribute the team gains/losses to players based on their ELO but the idea of net positive ELO sounds absurd

 

[vinigarcia87]

Gotta say... Only got good TG's since this change.
Fair games with people on the same level.
Finally good news on ranked :D