Multiplayer Forums

Hi everybody,

I am back to AoC after almost a year of inactivity. Thanks for the warm comeback receptivity!

Last year I helped writing the code for the rating system. I wish to better understand the failings of the rating system, and why it had to be reset in the last few months. I also wish to understand what is the so called "volatility exploit".

In the last rating system volatility always decreased with the number of played games. The idea was that the smaller the volatility the "better" we knew about the player true strengh. Therefore, the player would have a more "stable" rating, and would no longer have wild rating variations.

There was also a feature in which a player, when "leaving" in the middle of the game, could reset its own volatility, but would lose 15% of its rating points. I heard that some players had purposefully abandomned games only to increase their own volatilities to 1000 again. It was well know that such behavior was possible, but it never bothered me much, as I felt it was definitively not worth it.

A few months ago Lavi posted a screenshot of the Top 100:

viewtopic.php?f=47&t=5082

As far as I can see, there was no problem at the time. Is that correct? The most active players had ~400 volatility, which means their ratings updated only (0.4)^2 = 16% as fast as a new player. So, while a begginer would get +200 rating for a good game, and -200 for a bad game, such an active player would only get +32 for a good game and -32 for a bad one. Was that a problem? I heard from some players that volatility went way too low, and ratings were not updating anymore. Does +20 rating for a good game counts as "not updating anymore"? In the chess world, gradmasters playing against an even opponent only get +5 rating change from wins and -5 from loses. Of course I planned ratings to change more in a casual, slow-paced game such as AoC. However, I didn't think small volatility would be such a burden.

Also, I can't understand why would players chose to "abandon games" for increase in volatility. If the penalty is 15% rating, a 2800-rated player, then close to the top, would loose a whooping 420 points, and go down way back to 2420. I can't see why any player would do that, unless they don't care for their ratings or think theirs are way too low already. And if that is the case, it would be as if they had created a new account...

Is there anything I can do to help in regard to the new rating system?

PS: The new "different types of ratings" are very nice! It is a good thing for us to have separate team, blitz and slow games.

I'm guessing they left games, so that they could reset their rating in a way. Then, try again at winning and possibly getting a higher rank than was easily possible before purposely lowering rating.

Yes, I imagined that. But why was it a problem? Was it really an unfair advantage, or was it only that some players believed so? Did the top players in the ranking use such trick? Why was a rating reset necessary?

Not sure, the only person I know of who did it has quit the game. Don't know who you could ask.

hi,

a top player had over 3k points with low volatility and ever since the general field of players had gained relatively high ratings over time (without improving that much), it would make sense to disjoin games in order to lower the volatility and start anew.

high volatility plus very good results against average players with rather high ratings meant it was easy to surpass others at the top.

it also meant that the ranking was most certainly not accurate which is why it needed work done.

I can't answer most of your questions, because all the math was and still is way over my head. However, the above scenario actually became became very well known and passed around. As far as it not being worth it, again, I can't say either way. But for whatever reason, people thought it was.