Forged Alliance Forever

Re: From 1k to 2k after one game

2018-06-18T21:14:44+02:00

Someone clearly has no idea of how the system here works statistically

I only have 10 minutes so lets begin.

So basically FAF has a true skill system that is based on Gaussian distributions. They used to allow you to actually see the picture but now they just give you your mean and standard deviation.
See link on Gaussian distributions
http://hyperphysics.phy-astr.gsu.edu/hb ... aufcn.html

Your "rating" is actually your mean minus 3 times your standard deviation, as you play games both of these values get adjusted according to how you play during the game.

The trueskill system has some more variables under the hood to make these adjustments more accurate to how the system believes you should be rated. One of these takes how well you did in your last games into account.

so if a player is 1045 "rated" there mean is 2500 and there standard deviation is at 485 and they play that (one game after a series of wins) that completely cements there rating as 2500 against players that also have means that are 2000+ then there stand deviation may fall by a couple hundred. So yeah totally possible in circumstances where everyone in the game is a new player that has won a couple of there first 5 games.

The thing is that the trueskill system works really well at getting you to the correct rating in comparison to the players you are playing against. In general players tend to play in there rating bracket so it doesn't always come out correctly (and can vary widely when the there are only very new players)

Statistics: Posted by Freedomfighter — 18 Jun 2018, 21:14

Re: From 1k to 2k after one game

2018-06-11T18:33:04+02:00

replay id pls or at least your faf acc name

Statistics: Posted by PhilipJFry — 11 Jun 2018, 18:33

From 1k to 2k after one game

2018-06-11T18:30:36+02:00

I'm convinced that the rating system needs to be improved after seeing someone getting 1k points from a 3v3 victory against 2 big noobs and one half decent player.
There are players who solo 8 other players and get the usual reward after everyone else quit, so this system clearly is either not implemented correctly or the average microsoft developer (or copycat dude) messed it up.

I'm not going to expect anything, as there weren't any relevant changes in the past regarding to this matter, but someone should really try to revise that rubbish and use more inputs plus complex graphs or at least adapt to the latest standard. The game count might be a inversive exponential factor, but this is too much.

Sidenote: There are more recent versions of TrueSkill (2.0 as of 2018) and RNN algos.

Statistics: Posted by megaboi — 11 Jun 2018, 18:30