TruSkill is harsh

[FunkOff]

I've been recording my score for awhile now, and I have to say, TruSkill is a VERY hard rating system and may be ill-suited to FA, particularly in non-1v1 matches.

Specifically, last game I played I lost 21 rating for, which was an 8-man FFA where I was one of four players to score -1 in the Game. Further, I've won multiple 6-man FFAs were the total rating gain was less than 1. So, so I have to win 30 FFAs just to off-set a single FFA loss?

I think TruSkill was designed with first person shooters in mind, where in a FFA environment, one's K/D is a good reflection of one's skill. In FA FFAs, however, players will gang up on players with higher rating, or Rage quit to deny kills, and this makes accurate scores difficult to discern.

For example, on a 6 man FFA awhile ago, I placed second (I was second-to-last to die) and I lost 18 rating.

Winning a team game or FFA never generates more than 4 points, whereas losses usually incur around 20.


All I'm saying is, winning half my 6 man FFAs should not get me a net loss in score :-p

[Raging_Squirrel]

It's fine. You won't gain anything valuable for winning vs low rated players, though they will gain a lot winning vs you.
I heard you cheated on AI to increase your rating, so it might be true after all. And now the system puts you where you truly belong. Of course not without complaints

And considering that FAF cannot deal with FFA yet, and it have been said by zep many times that FA's game reporting system is fucked up and needs a rework, this thread is complete nonsense.

[FunkOff]

I heard you cheated on AI to increase your rating,

I've played games with noobs against AI, but I don't think that affects rating. XD

But yeah, according to the truskill website, FFA results are worth far more than team game results, and a loss in an 8 man FFA is the effective equivalent of 20 losses in a 4v4, even though FFAs are much, much easier to lose.

[Ze_PilOt]

Trueskill is pretty accurate at evaluating teams games.

The fact that you lose a lot and gain a few is only because the system expect that you win. If you win, as it was expected, your rating move barely (depending of a "confident" factor, yes, it make bets . The same if you lose.

With this system, you don't have inflation in the rating points, like on GPGNet. So it's totally normal that you don't increase your score a lot over time (if you do, it mean that you are becoming better, not that you play more).

For team game, it's the overall team that is rated, not you. your rating is then revaluated depending of the team result and overall score. The system doesn't care AT ALL if you die or not during that game (yes, a sacrifice can be profitable for your rating if it lead to a win for your team).
I repeat : Dying or not doesn't make a difference. Trueskill doesn't even know that you die. It only care if your team win or lose.

The problem is not trueskill, but that most of the game are not balanced, so one team will only gain few but can potentially lose a lot.
You can know the "bet" trueskill is making before the game starts. It's called "game quality", it's a percentage of a balance game.
100% mean that every team have equal chances or winning.
No game should be played with a game quality under 70%, for balance sake.
Most of the FAF games are rated below 50%.

So it's not a system problem, but a player problem : don't start a game that shouldn't happen if you care for rating.

FFA are rated correctly since last patch(it was a bug with random spawn, not trueskill).
For trueskill, it's a game with X teams of 1 player.

BUT it's true that FFA shouldn't count for the same as any other game mode due to the randomness factor.
But when does it become really random ? 1v1v1 are fairly balanced for exemple...
Also, if you doesn't lock the teams, the game shouldn't take it like a FFA..

[Ze_PilOt]

Oh, and AI doesn't count because they don't have any rating.

So except if you do 2 human vs 1 one AI and one human, it won't count (and in that case, it will count as a 2vs1)

[Styx]

You can know the "bet" trueskill is making before the game starts. It's called "game quality", it's a percentage of a balance game.
100% mean that every team have equal chances or winning.
No game should be played with a game quality under 70%, for balance sake.
Most of the FAF games are rated below 50%

How can we check that game quality percentage? Where can we see it ? i'm interested

[FunkOff]

Trueskill is pretty accurate at evaluating teams games.

Yeah. I'm not complaining about team games with 2 teams.

BUT it's true that FFA shouldn't count for the same as any other game mode due to the randomness factor.
But when does it become really random ? 1v1v1 are fairly balanced for exemple...
Also, if you doesn't lock the teams, the game shouldn't take it like a FFA..

FFA doesn't just count the same as other game modes, it counts for MORE.

In the context of FPS, this makes a lot of sense. After all, in FPS, if you die, you respawn and can kill more people. Also, there's no such thing as a snow-ball effect... 30 seconds into the game versus 30 minutes into the game, everybody has the same guns, the same hit points, the same potential for a skill difference to make one guy rise to the top and have a high kill/death. That's not true in FA FFA, though, if two high-skill players fight each other, a third low skill player, who is left to tech and eco, will be put into a position which, despite his lesser skill, he will be able to win just because of all the extra units/upgrades/resources that he has.

Also zep, you say it doens't matter whether you die or not, and Truskill doesn't see that, but Truskill DOES see your final game score, and if you die, you get -1. If it's an 8 man FFA and you have opponents who rage quit before you kill them and you are not ultimately victorious, the winner will have +5 but everybody else, even you if you were only the last to die, will have -1 and tie for last-place, and thus take a huge hit in rating because, according to trueskill, even though I was expected to place beneath Zock (who, in this example, ultimately won), I was not expected to place on the same level as huan36 (rating -277) and therefore my rating suffers immensely.

[FunkOff]

You can know the "bet" trueskill is making before the game starts. It's called "game quality", it's a percentage of a balance game.
100% mean that every team have equal chances or winning.
No game should be played with a game quality under 70%, for balance sake.
Most of the FAF games are rated below 50%

How can we check that game quality percentage? Where can we see it ? i'm interested

Yes, I also think FAF should post the game quality in the lobby... and give some indication which is the higher skill team.

[Ze_PilOt]

Some precisions, so new players won't be confuse, even if you are right :

FFA count for more because they are more teams (even it's only 1 person in them), so it can evaluate more precisely the performance between each team. It's not that it count for more, but that it can retrieve more informations from it.
But a 2vs2vs2vs2 will have the same impact as a 4 FFA.

Trueskill does see your score, but your team score, not your individual score. But indeed, in a FFA, it will be the same.
Also, it doesn't matter if you have (and all others) -1 or +4 if the winner has +5. At the end, what trueskill really see is the final order of the players, from top to bottom. Fact is that I order the players by the score given by FA (the score you see).

Indeed, FFA shouldn't matter much in your rating. It will be changed (remember : beta), but the problem is how to detect FFA.

That lead to the final point :
You can get the game quality by typing "get quality" in the FA lobby.
You can also type "get teams" to have the most balanced teams possible.
But it won't work because of auto-team :
Before the game starts, every games will be seen as FFA. I need to rethink the way autoteam send infos to the server in order to make it works again.

Also, the FA lobby can send infos to the server, but the opposite is not true. I have no way to display the game quality inside the lobby.

[Styx]

It would be great if in deed we could easily get an idea on how much teams are balanced or not.

If you need help for the "rethink", I'm in