[BLOG] Age of Empires II: Definitive Edition — Updates to Ranked Team Game ELO Acquisition

Updates to Ranked Team Game ELO Acquisition


Hello Community!

A major part of online play is online matchmaking. How the game decides who plays together can make or break the fun and challenging experience Age of Empires II: Definitive Edition players have come to expect from competitive matches.

That’s why today we’re making some changes to the way ranked ELO works in Team Games (TG) in Age II: DE. This change is specific to how ELO is awarded after team games, but we want to be clear: this is just the first of several changes we’re investigating to improve the competitive experience. Your feedback, as always, is vital, and we will continue to listen to your voice on our forums, Discord, and social channels to find new ways to give you great matches.

We’ve been tracking the issues you’re seeing in imbalanced Ranked Team Games, which has its root in how ELO wins/losses are currently calculated and distributed. The system can get pretty complicated, and an undesired ELO distribution can be more or less of a problem depending on the makeup in a given match. Part of how we wanted to address this was by simplifying both the solution and the conversation–please view the blog for full details and specifics about this change and how the new system will work, which will be implemented in today’s server maintenance:

[READ MORE]

10 Likes

Thank you all for listening

Since i made the big thread about this subject (Analyses of the ratings - Spotting the issues) i have to react to this news.

Great to see changes on this subject. It is really needed. I read it is part of the server maintainance today?! That is every greater!

You made a good explanation of what happened in the calculation and how to fix the calculation. The fix is conform my expections. Taking the average instead of the max will mainly fix this issue. So thanks for listening.

I do have a little remark based of the following comment:

I think this is the hardest challenge in fixing the roots of the issue. After the server maintainance the new calculations wont lead to inflation. But the effects on the current ladders take a long time to normalize. It even has some effects on the 1v1 ladders, because of what is used as starting rating on the 1v1 ladder if you already have a TG rating. With EW coming to ranked, it will even have some effect on the EW ladder that isnt started yet. So i think that those additional steps also needs to be taken quickly (at least before the EW ladders go live). But i do understand it is a tough discussion, since there isnt really an easy fix. All attempts in fixing the ladders, will have pros and cons. I can see three options as possible steps to help the process of normalizing:

  1. Full reset of the TG ladder. This will fix the issue, but everyone needs to play their 10 games again before getting a rating and everyone is losing his progress. I expect the first 10 games of everyone to be unbalanced for this reason and people dont really like losing there progress.

  2. Rescaling the ladder. By just dividing everyones rating by about 1.5 the average rating will become 1000 again. This way you get rid of the inflation, which is great if EW ladders are newly added to the game. But this fix wont really address the issue of people that now have a very high rating because they just played a lot. It will still take a long time for them to lower their rating.

  3. Full recalculation of the ladder based on the new algorithm. This will have the best results for everyone, but it requires the full game history of every account. I dont know how easy it is to recalculate everything for all ladders (since all ladders are connected by there starting elo).

So i wish you good luck wit taking additional steps to help with normalizing the ratings of everyone. Given the blog post i have good hope the devs will resolve this issue fully.

I also like these kind of blogs in which the dev team explains what they are doing. KUTGW!

5 Likes

I ask for a different ELO for premades. Like in starcraft2 or minion masters

But any way of addressing it (premades) like the soft penalty in CSGO could work.

1 Like

I have a question. Are you going to reset elos ? Because i am having some issue because of that elo boosting. Before i was losing 2 games = 10 points winning 1 game 13 point as a quick math, i was gaining elos even i was losing at the end of the game. Right now i have really bad elo and i want to reset to chase with better players.

I agree; I also expect a reset to TG elos. Call it season 2 or something. It’s 2021, it should be season 2 anyway.
1vs1 is fine though. Don’t touch that.

I hope you are not going with option 2 @RinTheYordle. Otherwise this fix would be a joke. Ideally option 3 could be used. If you no longer have the full history, go with option 1.

Option 3 seems like it won’t work especially since those games regardless of each player elo were won or lost would increase or decrease elo pretty much the same - it would not fix the issue because those games were imbalanced as is. Better to reset the entire thing and run with it

I see your point. A lot of matches were based on incorrect ELO so they shouldn’t happen in the first place. Reset is the best solution.

2 Likes

It will be more accurate then the current ladder, while having the benefit of keeping your progress. If the winning team was on paper also the best, then the Elo gain / lose would be minimal for these kind of games.

Resetting will also have my preference, but the devs have to make their own trade off.

1 Like

I also think that the inflation that already exists would have to be eliminated.

Proposal number 3 would certainly be the best - but I fear that would not be possible.

Proposal number 1 would be radical. Considering that the elo system is actually only used to generate fair and balanced games, it’s quite ok though. However, a lot of players will still complain if they lose their “hard-earned” rank.

If you don’t want the latter (in order not to scare away some players), proposal number 2 would be a still acceptable compromise. Sure, at first the elos are still not according to the skills and it takes months until they are. But day by day and week by week it would get better. And at least new 1v1 elos or other elos would not be affected by it anymore.
By the way, you should be able to calculate the factor by which the Elos are divided relatively accurately: If you add up all TG elos of all players and divide by the number of all players with TG elo, you get an average elo. This average elo should then be set in relation to the starting elo of 1000 (example: average elo is 1600 - then all elos would have to be divided by 1.6).

Whichever way: the main thing is that the existing inflation is also eliminated. Even if the root of the trouble may be eliminated. Good that this has now apparently happened.

Elos shouldn’t be divided. Why?

An elo difference of 100 means that there is a 3:1 bias. Division would destroy the difference. I liked SoTL’s approach of rather subtracting the difference of present average elo and required average elo.

Except it wont fix anything…

Currently we have people who are 2,5k, just because they played a lot in the inflated system, while they still preform pretty badly. You can spot these players by looking for a winrate far below 50%. For balances teams these players have to drop in their elo. On the other side you also have other players at 2.5k with a winrate far above 50%. They are much better players and may even have to move up the ladder.

This is the main issue at the TG ladder atm and this needs to be addressed. The solution from SotL doesnt address this at all. It only moves the average back to 1000 over a timeperiod of a year, but people that have an inflated elo still have to drop manually and that take ages. It takes months, maybe even years, before this will normalized manually. It will go to slowly to be effective.

I see his solution as a kind of variant of my #2 solution, in the sense of having the same drawbacks. From my 3 possible solution i think this is my worst, so i dont hope the devs will go for this approach.

The solution of SotL even doesnt have the benefit of fixing the ladder for EW games. Therefore the approach of SotL isnt really a good solution.

https://ratings.aoe2.se/

The median is about 1500. That is why i used 1.5 in my example. I think using the median instead of the average would be a better approach. This way 50% of the players is below 1000 and 50% is above 1000.

But i do hope the devs wont go for solution 2, because it doesnt really clean the ladder. If only reshapes the curve of the ratings, but many players have still an inflated elo. See also this full post.

Recalculation the full ladder based on the full history or just a full reset of the ladder seems to be the only options to consider. The reset has my preference, but i have no idea if the devs dare to take that step, since it will whipe all progress of all players, which might result in some hate from the player base. Personnally i dont really mind. For me elo is just a number which is used to match me against equally skilled opponents. I dont really care about the numbers at all. But some people do care.

1 Like

I watched the video of Spirit of the Law. Well explained and basically I think SOTL is also much better at math than me. But I’m not so sure that subtracting 500 elo is better than dividing as suggested.

This is because the elo gap, which determines the odds of winning, is also affected by inflation! So it is not true that a player who is 500 TG elo higher currently has a 95% chance of winning (even if you assume that both players have the same number of games and the player with the higher elo is therefore actually better and has not simply played more).

By the way, this would also lead to strange results:
A player with 3500 TG elo (ok, currently there is only one) would end up with 3000 TG elo - an elo that nobody should actually reach (compare 1v1 ladder).
Moreover, players with 500 or less TG elo would have to get a negative TG elo - which is actually not possible as far as I know. If these players fall to 0, inflation would still exist (albeit less).
Finally, a new player who may currently have a TG elo of 1000 will drop back to 500. This would be very stupid for him and he will rather make a new account than continue playing with his existing one (you have the problem with dividing as well, but not quite as much).

Perhaps I also have a thought error here. But if not, I think dividing is better than subtracting. But probably a complete reset is actually the best…

Yeah, he did a great job in explaining the issue.

But in his analyses he forgot something important. You already mentioned it:

He didnt looked at the mess on the team ladder, which is the result of the inflation. So these theorically percentages based on the theory dont really apply in practice as result of the inflation. That is the major flaw in his analyses. As result his suggestion isnt really a good suggestion.

He suggested to not subtract everything at once, but do it by small amounts every month. Like subtract 50 every month until the average / median is back around 1000k. Given this approach, you dont really have to worry of new players go from 1000 elo to 500 elo or inflation from people whos rating drop under 0 and thus is set to 0 again. The proces of deinflation just continues untill the average / median is back to 0.

You still have a much higher max rating for TGs and i do think that isnt really something what we want. Ideally all ladders have the same max rating.

In the end both substraction or dividing have kind of the same end result. But both dont really clean the ladder, so you still end up with lots of unbalanced games, due to the mess of the inflation. Both substraction and dividing wont solve that issue. As result both seem like easy to implement solutions, but with a bad outcome. In the end there isnt really much different between substraction or dividing. They seem equally bad to me.

The 1vs1 elo also needs some consideration. People who play team games first and then 1vs1, start with an elo higher than 1000. So that messes up things.

I do consider this as a good thing. A good player will start with an higher rating, a bad player will start lower. As result they are starting much closer to their true rating in most of the games.

But for this to work there is a assumption that needs to be forfilled: All ladders have the same kind of distribution. Currently we dont met this criteria, as result of the mess on the TG ladders. So that really needs attention. It would be great if the devs can clean up this mess somehow before we get the new EW ladders. To me that looks to be a requirement to make EW Ranked great.

So yes, they the devs need to be carefull with the connection between the different ladders.

1 Like

For 1vs1 ladder, a simple division would suffice. For team games, a reset is necessary

So will the DM elo be stored on the server or wiped out?

Why does 1v1 ladder need division? 1v1 elo calculation is correct except some players have a wrong starting elo. Recalculation of their elo is needed instead.