it's difficult because you might be an 8 on soldier but a 3 on scout. you could give a rating for each person's specific class. or you could just give everyone a general rating that they are assigned even if they off class.

have you considered an ELO-type system instead of stars?

Will people who you play with affect your ranking? (will they be able to rank you)

doesn't that open up the system for abuse? should be based off of W/L

speaking of abuse, your system better account for win-trading, as you're dealing with a small pug community and it could easily be arranged

@MR_SLIN: What is an ELO rating system?

http://lmgtfy.com/?q=Elo+rating+system

Elo is a name btw it's not an acronym

Is it just me or does it seem like someone posts a lobby alternate on these forums every other week and then we never hear about it again.

Otherwise, best of luck.

Elo is definitely what I'd use. You could give each player a "Scout Elo" / "Soldier Elo" etc. but is that really necessary?

yeah ur probably right a general elo for each player would be best

fuck elo, use this: http://en.wikipedia.org/wiki/TrueSkill

Well I'm not offering an opinion, just wondering how players would use such a system. Would having one general Elo rating for a player discourage that player from playing other classes? Is that a bad thing?

But, Elo is a 1 on 1 rating system isn't it? How would i implement it for 9v9 and 6v6?

Isn't Elo just between two players? How does that work 6v6

edit: you ninja

You know, I spent a lot of time thinking about this while thinking out my own lobby system before school picked up and I lost a lot of free time. It's pretty difficult to objectively gather data to rank people.

Obviously, you'd need to use data from logs to rate people. They would have to get points for the basic stuff: points for DDM, KA:D, etc. But there are several factors to consider, the biggest glaring issues being the medic and roamer.

Medics are a bit simpler; medics are rated based on deaths, heals dished out and ubers gotten/dropped. What would be even better is a solid heal spread, but that would be rough to calculate.

Roamers, on the other hand, are notorious for low damage and awful KA:Ds. How do you play roaming soldier in these games and not get a better rating?

Elo is great for things like chess and MGE, because they are 1v1 situations; you're in complete control over how well you do. But when you're on a team with 5 other people, Elo pretty much falls apart, since it's based on the rating of your opponents as well as whether you win or lose. If you have a high Elo, but get matched up with players a bit lower than you, and lose, you'll take a bigger hit than your teammates.

I think the biggest problem is that, even if you did find the perfect rating system, there simply won't be enough games to make things even. Someone who plays 100 games will most likely end up with a fantastic relative score, even if an invite player only puts in 10 games because they aren't all that beneficial. The rating system would potentially be useful for even matchmaking, but then, how often do you find more than 2-4 teams-worth of players looking for a PUG? It's hard to make this number useful when you only have a handful of choices for each position, compared to something like LoL or DotA where there are thousands of players at your level to be matched up with.

Anyway, the point I'm trying to make is that a rating seems like a great idea, but really isn't all that practical to implement in a meaningful way. If you were to do something, Elo is the obvious choice, if only for its simplicity and track record. Just don't make these numbers public or so help me.

When I made an Elo system, I did team average and logarithmic regression of point change as you moved away from the team average. It won't conserve the number of points in the system though, but it shouldn't matter too much (since it comes with playing more anyways).

I made a test system, but my K-factor is probably a bit too high (I think around 30). Columns for T1, T2, and score. I really don't like where it is, but it works if the games are close (when everyone is within 300 elo of each other). It's not meant for huge mismatch of skill level.
http://tfqueue.com/home/test

@synchro: Thanks for the input! I agree with you on a lot of these things.
I can definitely use this. I am no longer clueless!

Why would you like to make this info private and not public?

I always thought Damage per death was the best indicator of skill. Of course Elo would be better though.

Edit: And like Synchro said, you would have to compare roamers to only roamers and medics to only medics and so on. Each person would need 5 Elos, and then maybe a 6th for off-classing. Not sure if that's possible though.

Edit2: You could also take Elo from comparing just both roamers to each other (damage/death). Damage per death seems like the best way to rate all classes besides medic. And in the case of scout you would compare 4 players instead of 2.

Medic's are all in ELO hell then.

Synchro raises good points, so I feel the best way to do it is simply if you win, you get x points, if you lose, you get -x points. This is similar to the way League of Legends does it (it has a Elo rating for individuals, then averages them when teams come together), which also has defined roles (the support similar to the medic or roamer in this case). I think it's really the only way to do it. Then it just comes down to how many points do you win/lose? This should be based on the difference between the average rating of the two teams.

This will obviously have issues itself. LoL players constantly talk about "Elo hell", a low Elo ranking you can't escape no matter how good you are. In the end, one person cannot carry a team to victory (this will be even worse in HL), so it can be difficult to increase your Elo if you get a bad start. However all rating systems will have flaws, and perhaps you can have ways in the system to reward individual performances to overcome that issue.

I wrote a whole post arguing for and against elo, just to realize that tf2 doesn't need a ranking system especially for individual players. tf2 is a team based game, and it has a very small and tight-knit community that generally know the skill levels of other players. Something like this is not needed in my opinion.

The whole point of this would be to expand the community so you wouldn't be able to know all the members.

The thing is, for a class like Pyro where your job is to be a medpet how are you going to count ranks?

Well i'm thinking, as a pyro, a Spy pick could count for more than a normal kill?

Brownymaster wins thread in my opinion. Talk to him about implementing statistical ranking.

I'd rather not :/ . My system sucks, and I've tried things like person vs team average without and rating decay without coming to a good conclusion.
Just use a random PHP library and modify it for a team input. Example one I checked out: https://github.com/tushardhoot/EloRating

I pug a lot at Crit Fail (http://hl.crit-fail.net/) and they show KA:D when teams are being drafted and it's honestly a decent metric of general ability if everyone plays their best. There is an error of probabl +- 0.3 where they could be equal but it's good enough for casual pugs. EDIT: The ratio is shown for each class too which is helpful

The only way to do it correctly is to have it 100% based on the W/L of your teams.

Look at HoN (heroes of newerth). They use an ELO like system in a 5v5 teamgame pretty effectively in a game where they are effectively 50+ different characters with different roles. Ultimately the goal is for your team to win the game, not to play to optimize your KA:D ratio. There's a lot of actions in TF2 that win games for your team that is not reflected in KA:D ratio.

ELO ratings never function well in team-based games except at the highest and lowest points

they do in HoN...it's just those first matches before the algorithm has enough data with which to form the correct conclusions that it's inaccurate. There's various ways you can help jump start that data set, i.e. starting players from invite at x rating, IM at y rating, everyone else at z rating etc

Pheesh is right, HoN has the best "public" Elo system IMO.
HoN has a great system. That was my goal, and I tried to copy their decay system. I should have just googled it, because it's apparently public:
http://forums.heroesofnewerth.com/showthread.php?37833-PSR-Formulas