Analyses of the ratings - Spotting the issues

I guess such happens only when one wants to try some troll strats :slight_smile:

That would counter the way the smurfs are abusing the system. However, my idea to calculate team Elo based on an exponential average does the same thing, except that it would also lead to better matchmaking if the rating differences are smaller, instead of having an arbitrary cutoff point.

As an example, if āˆ† = 500, then the team rating cannot be more than 500 points below the highest rated player in a 2v2.

1 Like

You can add some floor on different ways. My example was just a quick and easy approach. You can make it some more advanced. I just played around with your suggestion and in practise it will be about the same outcome. The biggest difference seems to happen within the 500 elo range. If will boost the team elo in these cases as well.

I think it is up to the devs to decised of quick and easy is already good enough or that they want some more advanced calculations, based on the exponential scale. Therefore they just have to analyse the data and see what option has the best fit.

please stop posting before some dev picks up on your horrible ā€œideasā€. What would be the purpose of removing tg elo, except making the whole tg playerbase quit the game?

you must be kidding, devs dont give a damn here, you worry too much

What is the opinion of people of changing the rating from an elo system to something else?

Microsoft already have its own TrueSkill algorithm. Glicko-2 is also another well known algorithm.

Based on the theory both are an improvement over the current calculation. Both are also made to work better for team games as well. They might even help with smurfs, since their initial rating will grow a lot faster (i image).

2 Likes

we need ratings for entire teams.

when teams havenā€™t had history, only then do we need to fallback to using some alternate predictor.

the whole incremental rating adjustment concept by a series of +4s and -4s is silly. it takes hundreds and hundreds of games to move teams to where they belong.

all the teams with 70% (some even at 80-90%) win rate indicates a complete failure of the matchmaking system to deliver quality games. after hundreds of games they are still being handed easy matches they are 99% to win instead of fair games

there should also be zero tolerance for games that have a large rating difference. a good rule is that if a game is more lopsided than 80%/20%, do not even launch it

2 Likes

Could you guys elaborate on the ratings for entire teams thing? What happens when you go solo queue and meet 3 other fellas youā€™ve never played with before, does your rating match you with people who just downloaded the game? Or maybe itā€™s impossible to queue without a full team? Am I forced to play with the same 3 guys over and over if I want accurate matchmaking, because playing with a different friend will set our team back to 1000 elo?

you can have a rating for each set of players and an individual rating. every game can update each value.

then you estimate a teamā€™s skill based on the data you have. if some/all of them have played a lot before, then give those ratings more weight. if they havenā€™t, give the individual rating more weight.

other games like starcraft 2 do something similar, but i donā€™t know what exact formulas they use

1 Like

We do this because TG Elo is wrong by design, so the solution is not to kill TG Elo, but to fix it.
If TG elo would be reliable we wouldnā€™t give 1v1 Elo that much importance when playing team games.
In order to fix TG elo they just need to restart the rank for every player and prevent players with 1k elo difference from playing together.

bump! Please spot the new issues.

I think there are two main issues:

  1. We still have the mess of the old calculation, so i suggest a ladder reset to clean this up.
  2. Smurfs exploit the system. Smurfing is bad in general. This point can be solved in different ways, like limit smurfing, but also changing the elo calculation so it has less impact on the outcome.

I think these two issues are the current main issues that needs to be fixed.

1 Like

Just a small sentence in the patch notes. It isnt even under ongoing investigations, but it seems like they are looking for a way to clean up the mess from the TG elo and also fixing the smurfing exploit. I really hope that they are doing this quickly, since all the smurfs are killing the TG ladder at this moment.

3 Likes

Anyone has heard something about when the devs will finally fix the team game ratings?

The solution they have implemented would be good, if it hadnt the following drawbacks:

  1. They still have to clean up their mess of the old calculation.
  2. The new calculation is very exploitable by smurfs, so we see many smurfs on the ladder.

They updated the calculation 2-3 months ago. They even said they will still look into this. So i really hope these issues are taken care off.

1 Like

I think that patch note was merely to calm the waters and give us the impression that they are working into it, but i have my doubts, resetting TG ladder is task that takes like 20 minutes, porting the 1x1 to the TG ranks to start with something instead of 0 just another 20 mins, that would fix the point trading and the big rank disparity that is plaguing now the ladder.

Also fixing or changing the MM algorithm to make better matches is something that should only take few hours, its been a month soā€¦i really canā€™t understand why are they taking so long to fix the basics.

If they fix the ladder first, then the alt f4 would decrease automatically since people like me wont get matched with 1k 1x1 players all the time on team games.

What do you all think of changing the team elo math to something like this - A proposed team elo solution

with your system, one of the old problems would come back. I often play with friends of different skills. With your system, we would at the end get the same rating. If then we change team, some of us become smurfs and the other are overrated.

Please read our suggestions (we suggested at least 2 ways to solve your problem without creating new ones).

The best suggestions we had is to change the way the average rating of the team is calculated, where the highest rating of the team have more impact on the average, and then limiting the smurf impact.
With the exponentiel scale average and a Delta of 500, a 4v4 team with a 3500 player could NEVER have a rating below 2500, even with 3 0elo players with him.
we could even think of a lower value of the Delta, to discourage it more.

i canā€™t see any reason not to use this suggestion.

1 Like

I do think you kind of underestimate the time. First they have to decide on how they will fix the issue, then they have to work on the fix. Then they should also reset the ladder (should be part of the fix, like you suggests). But they should also test the solution. They probably have some OTAP, so they have to pass all steps during the process before it can go in production. All in all i think it should take a bit more then just a few hours. But in the end i agree with your conclusion: The devs should already have had more then enough time to fix the issues and i also have no idea why it takes so long for this being fixed. I already made this thread in May 2020, 15 months ago and this wasnt the first thread about this subject. So we are already waiting for about 18 monhts before the devs finally fully fix this system.

I posted my opinion already in that thread, but i should reply here too. I dont think your solution will fix anything. Smurfs will still be an issue. They only need to replace the smurf account a bit more quickly. A solution in which the TG rating is not just the average of all TG ratings, but more skewed to the highest player should be a much better solution. The matching and the elo gains should both be based on this new TG rating for a team. They devs should also reset (and maybe port the 1v1 ratings as start) the TG ladder, to clean up the current mess on the ladder.

I have seen multiple suggestions with kind of the same outcome: More weight towards the higher rated player. Exponential scale is just one of them. I am kind of indifferent between the multiple suggestions, as the end result will be more weight towards the higher rated player. That really should be the goal, because that would really limit smurfing. If a 3500 rated players with 1000 elo allies still count as something like 3000 TG elo and is matched accordingly, then it isnt worthwhile to smurf anymore.

So i agree with you that we need something like this to stop the smurfing that is currently ruining the TG ladder. The other issue is the current mess of the current and previous calculation The devs really need to clean this up. The most easy solution would be to let everyone starts again, but then it will take some months again to settle. Other option is to just recalculate all the games of the past. But the match making was pretty bad. So i am not really sure if that will really fix the issue. Other option is to port the 1v1 ratings to the TG ladder as starting value. We always had this connection in the past. If you start fresh on a ladder, but you already are on a different ladder, then your rating on the other ladders is used as base instead of the 1k. With the release of EW they removed this feature. But i think that the devs could put that feature back into the game after a good fix. This way the 1v1 RM rating will be used as base for the TG ratings. That sound to me as the best solution. The devs already have this code. Only thing they have to do is put it back into the game and reset all TG ratings.

2 Likes

Well i think they are going to release the fix with the upcoming patch+dlc in few days.

The team elo is currently broken. It actively incentivizes smurfing. A 2000 ELO player teaming up with 3 apparent 1000 ELO player gains a lot of ELO by doing so and can become a 2500 even if his/her skills do not merit that. Itā€™s destroying the fun on the TG ranked ladder. This needs urgent fixing.

Hereā€™s a simple and fair ELO distribution system that will not lead to inflation and result in more fair games. Say, itā€™s team A vs B and A wins.

Award teams A and B the ELO that is currently awarded based on avg ELO. Letā€™s say team A gets 10 ELO and team B gets -10 ELO. That much is fair and avoids inflation. Now we donā€™t distribute the ELO to the individual players equally. We distribute it in proportion to their strength vs the avg ELO of B!

So say A has 2 players a1 and a2. If a1 had defeated avg(B) ELO he would get 20 points because he is a weak player and a2 would get say 4 points because he is a strong player. Give a1 (20/24)*10 ELO and a2 (4/24)*10 ELO. You get the idea.

Give each player in the team ELO loss or gain proportional to what they would have gotten had they played against avg ELO of the other side. Distribute a teamā€™s ELO in proportion to relative strengths of the team members.

This way if someone teams up with smurfs, the smurfs get majority of the ELO and the non-smurfer gains very little every game. This removes a lot of the incentive of smurfing and everyone regresses to their average team skill.

This is more fair than the current system and just as easy to understand and hopefully implement.

(First post here, please be kind if I broke any etiquette.)

3 Likes