Game counted?

[Lykourgos]

QUOTE (mcwarren4 @ Sep 9 2006, 02:03 AM) With tighter constraints it wouldn't be so easy to exploit that flaw.

No. With tighter constraints fewer games would be counted and the stackers' ELOs would rise more slowly, defeating your own purpose.

Greator, how you got THAT out of Night's post I will never understand. Night made a comment about the behavior of a mathematical system- you took it as a statement of intent to stack?

[mcwarren4]

QUOTE (Lykourgos @ Sep 8 2006, 02:01 PM) No. With tighter constraints fewer games would be counted and the stackers' ELOs would rise more slowly, defeating your own purpose.

I respectfully disagree. The reason is the ELO gained or lost is more highly correlated to the probability of winning the closer to 50-50 you get. You don't gain enough ELO for winning a game where ELO says your chances of winning are .15. When you have a .40 chance of winning the ELO gain is more in line with the real probability that you might actually win.

So you have two possible solutions to correct the problem.

1. Tighten the constraints so that the only games that count are ones where ELO point adjustments are more highly correlated to the probability of winning.

2. Adjust the points gained/lost to be more reflective of the real probability of winning.

[FingerBang]

another option would be to LOWER the amount of points AVAILABLE.

If a stack is geater than 80% you get 1 point for a win. and the stacked against team gets 5
If the stack is at 70% you get 2 points stacked against loses 2
60% +5 lsoer -5
50-50 +/-15 points.

if you win against a stack

20% you get 30 points stacked team loses 30
30% you get 20 they lose 20
40 u get 10 they lose 10
and 50-50 same as above.

not sure if this sort of points can be implimented but sure would make things interesting in regards to stacking.

I feel that winning a severly stacked against game doesnt reward enough points, hell if you win with a 20%give the team 50 points, its a huge deal to win against a stack like that.


losing full points for not playing the full game is, im my opinion a huge steaming pile of crap. If i see a game thats been going for an hour say, i antistack and play for 6 minutes and we lose i lose full points but if we win i get a small % of win points? what kind of crap is that.

The % for a win or a loss should be the exact same. Even this doesnt correctly fix ELO but it is alot more fair. People are sitting out of games they want to play because there is the chance that the game will end 6 minutes after they join and they will lose ELO.


just some ideas..

[Greator_SST]

...The topic of this thread by mcwarren is such an accurate distillation of the main frustration with ELO ranking, thanks for posting, but the truth is, elo hasn't changed team joining behaviors at all and in the end, it doesn't matter at all. For me, the sooner the balance patch is implemented the better.

My only question about the balance patch is that 'theoretically' a person's ranking will only fluctuate up or down a certain amount until it self-corrects. So, theoretically, people will stay at a pretty constant rank forever if it works right, is that correct? Because once your rank rises, it will be counterbalanced by the balance and you'll drop down again.

But if that's so, doesn't it hold true for all ranks? That would mean that you're going to stay at whatever rank you start at (aside from the date of birth calculation). No one will progress up or down because the balance theoretically ensures that games are always balanced, ergo, you'll win as many as you lose. So intermediates and newbs will stay where they are as will vets.

Or is this competely wrong?

[Grimmwolf_GB]

No, Greator, you would not stay the same.
For example, both teams have 500 ELO rank points after a reset.

Team 1:
Hardcore vet 1
Hardcore vet 2
Hardcore vet 3
Hardcore vet 4
Hardcore vet 5
Hardcore vet 6

Team 2:
Voob 1
Voob 2
Voob 3
Voob 4
Voob 5
Voob 6

Both teams have the same ELO, because the ranks were set to 15 in the beginning. Now team 1 wins the game, causing all hardcore vets to go up a rank. The Voobs drop in rank

The next game has 3 Voobs and 3 Hardcore vets in each team. Both teams have 500 ELO ran points.

Now the ranks start to differentiate: What ever team wins, gets the ELO points. The team with better pilots gains ELO, the other loses ELO. The third game setup will look pretty different again.

After a while, the ELO system should show who is constantly winning and therefor a better player (higher rank) and thus should be balanced with a weaker. But perhaps that weaker pilot is also a good, yet ELO unknown player, or the ultimate vet outweighs the other player. He gains points again.

After a while, the ELO ranks should be stable though, because in the end that's what we want: even games were teams really have the same chance of winning (skill wise).

[mcwarren4]

Greator, that is where measuring true skill comes in. If Night joins a team that has a 20% chance of winning, he is good enough to make a difference but probably not good enough to win the game for that team. If Night joins a team that and its probability of winning goes from say 35% to 40%, a pilot like Night has enough skill to have an impact on the outcome... primarily because he has a supporting cast of pilots that are reasonably competant relative to the other team, but the end result is Night was able to make that team a better team. So what ends up happening is Night's rank will go up until the impact of him joining a team either 1. throws the balance out of whack (which would prevent him from joining the team) or 2. each game he plays HE is only able to have enough impact to win 50% of the time, which would cause his ELO rank to stablize somewhere near that level.

Finger - lets say 1000 games in a row there is an 80/20 stack by mixing the players up each game. Under those conditions all ELO's continually rise.

If constraints aren't tightened then the next best thing I can come up with is to create a formula that combines predicted outcomes from ELO with real outcomes and distribute points accordingly:

If probability of win = .80, teams with .80 probability of winning wins in reality 20 times to 1 loss, then points awarded should be equal to 1/20th of points awarded for a 50-50 chance of winning.

If probability of win = .75, teams with .75 probability of winning wins in reality 10 times to 1 loss, then points awarded should be equal to 1/10th of points awarded for a 50-50 chance of winning.

If probability of win = .65, teams with .65 probability of winning wins in reality 4 times to 1 loss, then points awarded should be equal to 1/4th of points awarded for a 50-50 chance of winning.

So in those examples, if a team wins a game where the probability of winning is equal to the probability of losing and is awarded 20 ELO points for winning the following would happen:

Team with .8 probability of winning wins, 1 point is gained.
Team with .8 probability of winning loses, 20 points are lost
Team with .75 probability of winning wins, 2 points gained
Team with .75 probability of winning loses, 18 points lost
Team with .65 probability of winning wins, 5 points gained
Team with .65 probability of winning loses, 15 points lost

[Greator_SST]

QUOTE (Grimmwolf_GB @ Sep 8 2006, 03:38 PM) ...After a while, the ELO ranks should be stable though, because in the end that's what we want: even games were teams really have the same chance of winning (skill wise).

I thought that's exactly what I was saying, the elo ranks will stabilize because the system will balance teams, giving every team an even chance of winning.

But let me try and explain it again (I'm probably wrong, but I can't see where I'm wrong). If the system really gives a 50/50 chance of winning, if I'm an intermediate, I'm going to win 50% of my games and lose 50% of my games. If I'm a veteran, I'll win 50% of my games and lose 50% of my games. In both cases the balancing system does its job, but in both cases, the rank stays the same.

I do understand that theoretically over a long period of time, the various elo points awarded for the minute differences in team elo might provide an accurate rank. But I can't visualize that. If I'm the worst player in allegiance, I'm still going to win around 50% of my games and lose around 50% of my games. If I'm the best player in allegiance, the exact same thing should happen, shouldn't it? Which is to say, there is little or no change in rank.

What am I missing here?

[Edit: Oh, and yes, mcwarren, although I'm not a statistician, I kind of like that points allocation table]

[Grim_Reaper_4u]

Why go through all the trouble of adjusting ELo for game imbalance :

a) do not allow games to start with more than 60/40 imbalance
do not allow games to get imbalanced when they run

In short : implement the balance button as a non-optional feature or if you feel like stacking ----> turn auto-balance off = turn stats off (better yet : piss of to a non-ranked server and go beat up on newbs there and don't waste our time on the main server /mad.gif" style="vertical-align:middle" emoid=":mad:" border="0" alt="mad.gif" /> )

[mcwarren4]

I posted a poll in the FAZ Development Forum regarding constraints. Take the discussion there.

Linky...

http://www.freeallegiance.org/forums/index...c=24318&st=

[Grimmwolf_GB]

QUOTE (Greator_SST @ Sep 8 2006, 10:37 PM) IIf I'm the worst player in allegiance, I'm still going to win around 50% of my games and lose around 50% of my games. If I'm the best player in allegiance, the exact same thing should happen, shouldn't it? Which is to say, there is little or no change in rank.

No, if you are the worst player in Allegiance, the system will initially balance it to a 50/50 chance. BUT: you suck. You chances are lower than 50 just because of YOU. Your team will lose more often, thus you will lose more elo than you will gain. The next games will reflect that imbalance. After a long time, you should have a lousy rank.

Same with you being the best player in Allegiance. ELO balance you on a mediocre rank and the game has 50/50 chance. BUT: you rock. You defend all the miners and kill the enemy economy. You are a huge part in winning the games for your team. Teams win more often than predicted, because you rock. After a long time, you should have a very high rank.