Forums

Wasn't the normal, was some random vertex on the mesh:  

@wowgetoffyourcellphone  

If you say it's common, then so be it, but from other situations I've seen that individual ratings take some time to stabilise, that would be my only concern... that for most of the "season" ratings would be quite inaccurate. My proposal (a mix of the new FIFA ranking system with a softened version of the old one, which had a harsh decay) would soften that reset. EDIT: Another reason I see decay as necessary is that players that have been inactive for too long would have been playing a somewhat different game… this is not as constant as chess, and even in chess there is an “inactive” category, where for example Kasparov is placed (still being ranked 2nd overall).

@kheeva @Thalatta Implementing some kind of rating decay or reset seems quite appropriate to me. There are players who have had the same rating without playing a single 1v1 rated game for many years . Perhaps instead of getting tangled up in intricate calculations, “seasons” could be implemented. Basically, with each new release, all ratings would be reset and everyone would start over. It’s a fairly common mechanism. 

Ok, I think I get what you are trying to write here. You think I have never seen a solution like yours, that I've only seen the solutions I’ve mentioned, and that that’s not an argument to dismiss your solution. Well, you are wrong, I have seen solutions like yours, that’s why I say they are poorly implemented, because they show a lack of understanding on how an Elo ranking works, and the solutions I proposed solve that. The issue is, the rating is related to the winning probability, although this depends on the scale chosen. In chess (maybe 0 A.D. is the same), being ranked 400 over someone means the odds to win are 10:1, while being ranked 800 over someone is not two times better, but ten times better, the odds being 100:1. That’s why, when wanting to implement a decay, if every week you subtract different numbers from different players, you break the rating differences, and the corresponding probabilities stop making any sense. Your “we can stop here. also can decrease maximum 300 elo and make not lower 1200 that way” phrase just destroys the meaning of the Elo ratings. I mentioned the zero-sum game representation because there’s a mapping between that and the probabilities, which you mess up with the subtraction, as they are related to relative differences. The only way I see to solve this issue with your method is to subtract a given number from everyone every time, but one also has to avoid negative Elo rankings, so this number has to be obtained as a fraction of the lowest Elo rating in the ranking, which could eventually return smaller and smaller numbers with time, defeating the purpose of the decay. Thus, subtraction is a bad solution. If, on the other hand, you implement the decay as I said, reducing the influence of older games equally for everyone, the probabilities are affected correspondingly, the decay behaves always the same, and Elo ratings behave normally. Is this with “solid idea”? Or is there something still unclear?