Analyses of the ratings - Spotting the issues

To be fair to the devs, in the latest change not only dodgers are punished, but early resigners as well. This makes it harder to smurf. Of course, it can’t be said to really be a ‘solution’ for surfing, either, but it’s something!

I am glad with the Alt F4 fix. Of course, if you were e.g. always playing Arabia and now have to play an Arena game now and then, these Arena games will be unbalanced until you figure out the map. That is not ideal. Previously, though, if you wanted to switch it up and e.g. unban Arena to try it out, you would get stomped because you were inexperienced at the map. This discouraged players from trying out maps they were not already familiar with and made it so that more people wanted to Alt F4.

In the same post where I suggested a teamgame system where the exponential average, rather than the linear average, was taken into account in matchmaking - which should definitely be implemented - I also alluded to how it should be possible to have seperate ratings for all different maps without players having to play placement matches for all of them. It could be interesting to implement something like that, too, though I doubt the devs will have sufficient resources and knowledge in their team to implement such a thing.

2 Likes

Already any news about a fix? Or is the game death now AoE4 is released?

Seems like no one even care about this issue anymore?!

Some people do care, some people even have solutions, but these people are not in charge …

This is my usual post to get attention for this subject…

Ratings are broken from the release and still not fixed after 2 years. The devs are really ignoring this issue, which is ashame.

1 Like

Another week without a fix for something that is broken since the release…

1 Like

I am sure they will adress something to it in the next patch

Any source for this claim? I really hope you are right.

I hope they will include the
Exponential Average
Proposal by @Mercy9545 finally.

The Elo system is a logarithmic projection of the strength distribution so the exponential average is exactly what we need to accurately represent the “real strength” of a team.

2 Likes

Will we get a fix for the TG ratings as Xmas present? That would be a great surprise!

1 Like

I agree that this thread is not easy to follow if you are not already involved in the discussion. However, I don’t mind @WoodsierCorn696 bumping this thread, and I especially don’t mind certain people referencing this post of mine now and then. :stuck_out_tongue:

It really shouldn’t be that expensive for them to contract out some bespoke mathematics work to solve these problems.

bro you don’t even need undergrad math to get things done.
company based programming training and highschool math is enough to fix matchmaking in theory.

in practice people refuse to work on this at all.

While it can be done with highschool math, most people don’t have the ability to correctly apply highschool math to problems like this. Those that can, go on to be much higher educated. When you employ them, you aren’t necessarily making use of the additional knowledge they have, it’s often just a way of knowing you have someone with more talent at applying the basics.

1 Like

They do not have to hire people for these maths … some people on this forum can do it for free, if they bother to ask !

I’d be happy to do the work for them, but not for free. If anyone with the right skills is willing to do the work for free, I suggest they contact the devs rather than expecting the devs to be aware that they have the skills and free time to do that. I’d suggest, though, that it would be a rare person who would both have the required skills and experience, and be willing to do it for free. More likely is that anyone considering doing it for free has never undertaken such work commercially and is underestimating the time required to do the job properly.

The “math” part of this is already 99% done here.

The “implementation part” does not require any skill that devs do not have.

1 Like

The reason we are where we are is because they just took a suggestion from this forum and implemented it without fully understanding the problem. The post you linked to is 1% of the work, it’s an idea that might be a good one. Analysing it properly (and potentially changing it as a result of that analysis) is the remaining 99% of the work.

Someone with only high school math can’t really do the job. It needs be someone with some more experience. The formulas in this thread might be easy to understand for them but until now all suggestions are made from a theoretical point of view. I would suggest to also run some simulations and test if the new model has a good fit. Or compare the fit of different models and decided based on these statistics which model will be the best. Give someone with only high school math the task to set this all up and test it appropriately seem to be a no go for me. Your really need to have more experience with math to do this well.

Programming is pretty close to math, so I expect the dev team would already someone who is suitable for this job, I would think.

The suggested log model also contains a parameter. I would suggest to set something up to monitor the outcome so you can change this value in time if needed. It might be starting at 400 and change to 500. Or the analysis shows it needs to depend on the average elo of a team, so they can monitor things like that as well.

Just picking an suggestion based on the theory without any practical testing seems a dumb idea.

1 Like

We already have a solutio from @Mercy9545 :
The exponential average for team elo instead of the arithmetic.

But ok let’s just pretend that never has been told.

ELO system is a logarithmic projection of the strength distribution, so we need to take the exponential average of it, otherwise we just get bs. (exponential average means the elo wil be projected back to non-logarithmic and from there projected to logarithmic again to stay in the elo system)

1 Like