cashto wrote:QUOTE (cashto @ Feb 5 2013, 06:29 PM) You said "lose less rank". That's exactly what AS does when you lose against a stack.
Of course wins get you rank. Wins are ultimately the only statistic that matters. I can bomb a lot of useless ops and kill miners long after the enemy has reached endgane tech but if it doesn't translate into a win, what good is it?
Even in a losing team in a game you can say that one player was better then an another, if they both have the same mu shu zu elo rank they will both take the same rank hit, my proposal is to use some metric so perhaps even in the losing team a good player can still increase rank or the very least lose less rank. In gameplay terms yes it might mean you can kill miners and ops even when the enemy team is almost certain to get victory but ask yourself is that really a bad thing to try and actually still fight instead of just giving up?
QUOTE I mean, let's say you get all the data you need. Let's say you get statistics that can distinguish podding a nan from chasing noob scouts through empty sectors, that can count the number of bullets you land on a bomb run and credits you accordingly for the kill. How do you weight the importance of events according to objective criteria? How do you calculate that killing a miner is worth n.nn times as many points as eyeing a bomb run? How do score someone who is a miner killer
par excellence but posts resigns and actively works to demoralize his team after only a couple of setbacks? How does the system figure out that the guy who got podded a zillion times, who never flew anything but a scout, and never left friendly sectors is the guy who won the game for you because he was sitting with the miners, kept them safe and mined you to victory? How do you, in short, write an algorithm to untangle the chaotic web of a million possible interactions that could happen in a game in order to scientifically proportion credit for the one discrete, undeniable result that you care about? It's impossible.
There's something to be said for feel-good points and XP points which correlate more with time-in-game than any skill factor, but never in a million years would you use that info to balance teams.
HSharp wrote:
Elo, Trueskill is poop for alleg
Says who? How do you know this?
[/quote]
First point is that without the data to play with I can't create an algorithm to play with. Just like Elo/Trueskill can't empirically be proven to be the best for Alleg I can't prove another algorithm to 'scientifically proportion credit for the one discrete undeniable result I care about'. What could be done though is much how people can rate players in team sports and that is by public opinion. It's hard to prove one footballer is better then another but you can watch a game and have an impression which one performed better and that would be some ideal ranking system, some AI that could watch all the games to apportion credit to the right players but I don't think that is possible.
Expert opinion could indicate if one algorithm is better then another much like peer review.
More metrics, more data, better algorithms. Without the data I cannot provide an algorithm, with it perhaps I (or someone else) could, or at least a better system then the current.
QUOTE That's because professional sports don't play pickup games. When the team composition is identical or near identical every game, Truskill is blind. Now if you randomized the players every game, Truskill is going to
very quickly notice when one side has statistically worse players than the other.[/quote]
It doesn't matter about team composition, it doesn't even have to be professionals, when experts watch a team sport they can identify which players played better. Trueskill might actually get a chance if teams were randomized every game, but they aren't so a system that works when taking stack into account might be better.
Also there are other games which do use XP to match-make, I could lose every game in War Thunder or Call of Duty and still reach high ranks, I quite like War Thunder's model which is a points based system in a game with team sizes of ~15 players, of course that is randomized and for all I know there might be some secret Elo type ranking but even when losing I'm not punished and in-fact rewarded depending on my skill. Obviously easier to tell a skilful player as it's rather simpler and there isn't a support role.
Who knows maybe the best ranking system is letting players vote on other players after the game ends. Without data I can't test @#(!.
