Free Allegiance Forums

Send Weed a handle of whiskey every time he logs in.

I'm not even good at the game, why do I have anything over a 15?

Maybe stacking should give you a huge rank, especially if you're a game changer, it should be up in the 35-50 range

20-22 just isn't enough, because you most certainly are not only twice as good as an average voob. Being able to pod 10 pilots in similar ships, all at once, is proof of that. Or commanding 10 times better than anyone else in the game...

DonKarnage wrote:QUOTE (DonKarnage @ Feb 21 2012, 07:47 AM) I'm not even good at the game, why do I have anything over a 15?

Because you created a new account and left the newbie phase of you old account behind you?

Problem

Newbie phase (or "Kick them while they're down"):
Mu goes down rapidly because they loose games (they don't know better). While they're learning the game, their sigma goes down, too. (See A1r(-1) on leaderboard.)

That means, people without looses in the beginning (returning vets, people who play longer than any ranking system, lucky newbies) end up with a higher rank, than people which learned the game the hard game.

Mu = "ELO" how are the changes to win a game (25 = 50/50)
Sigma = How assure is that Mu (starts with 8.33 and goes down with every game). The lower the Sigma, the smaller the changes on Mu.
Rank = (Mu - (Sigma x 3)) x 0.6







Solution

Don't count games, until:
a) People enable ranking system.
b) People reach the "magic number" of games (autoenable ranking system).
c) Enforcement enables ranking system.

We've been over all of this before but

Or the ranking system catches the fact that Weylin restarted his account, knows the game, and ranks him correctly.
The graph doesn't tell much given you don't know who the two players played with or against. ( not disagreeing or agreeing just noting )
Sigma can and does go up. It has been doing so for me.
Ranks converge rather quickly it has you in about 50 or so games
A sliding window won't greatly change your rank as you are STILL being compared to the rest of the community
A popularity poll is useful in the sense of getting a feel for those who feel strongly about the issue and nothing more
Ranks are not what we should be using but must given how the system had to be implemented ( Mu, Sigma, general NOAT, and autobalance)


Ok this entire topic has been a third rail in Allegiance since the day I logged onto the game.

I'm going to leave you with this

The current ranking system suffers because we depend on the players to implement balanced match play rather than the algorithm. This has been a similar knock on each and every ranking system that has been introduced to date. The first points based system was awful, grind based, and a Turing box. Alleg Age is useless and meant to massage egos rather then balance game play. Basic ELO based implementations ( like the last several versions ) are less accurate given the highly instable teams and problematic with regard to newbs ( You can see that it works well for the tournaments given squads are mostly stable teams ).

Many of the knocks on ranks have to do with something entirely different. There is no rewards for Allegiance other than game play experience and why the idea that Allegiance is a new version of Chess makes very good sense. Ranks are not meant to take forever to get to the point, make you feel good about yourself, or reward you for probing all game. They serve one purpose to promote fair and balanced match play. It is unavoidable that one will compare ranks, it is human nature and we aren't going to change that one. Given how ww've implemented the system to date it is unavoidable that we must display ranks in game. Which means after a wonderful evening of game play where yoiu wtf pwn the world your rank * crickets chirp * does $#@! all. Cause and this needs to be emphasised again, it is supposed to know your place in the community as fast and accurately as possible. Drop a thousand probes, pod Weed six times in a row, discover the cure for how to get them to shoot the $#@!ing nan... same rank. It makes people nuts and poor old ranks takes it in the shorts.

Can the present system be improved? * shrug * of course, Ive never seen anything perfect in my life, the database probably needs a tidy since no one has done much with it to the best of my knowledge in about three years, and it seems that some of the long timers ranks probably don't reflect their skills indicating there could be some tweaks. Basically we've never implemented correctly and that makes it easier to break it.

What I'd like to see implemented is:

Have a tidy with the database
A peak at the internals to see if the list of tweaks discussed makes sense
Implement Autobalance for AS on ACSS release to learn the best way forward with regard HOW to implement without disrupting game play on an individual level
Implement shiney things to reward hard work
Implement general NOAT
Enforce autobalanced game play for games that will count for statitics
Never display a rank in game again

Listen to the mindless complaining for the next four years.

* happy sigh *

I see this occuring in stages and COULD be implemented by the end of the year. Just need people to make it go


Thanks for listening.... and Allegiance community so much I'm going to hire Archie to give all of you a great big butt kiss
MrChaos

MrChaos wrote:QUOTE (MrChaos @ Feb 21 2012, 12:17 PM) Sigma can and does go up. It has been doing so for me.

So you wanna tell us that AllegSkill is broken?

As far I know sigma CAN'T GO UP by design...

pkk wrote:QUOTE (pkk @ Feb 21 2012, 06:34 AM) So you wanna tell us that AllegSkill is broken?

As far I know sigma CAN'T GO UP by design...

Then you are under a misconception brought on by someone writing something inaccurate. Probably fairer it was written for a general audience and not meant to discuss the nuances involved in the matter. The system has become less " confident" in my rank as Ive been winning games I should be losing which increases my Mu but also my Sigma.

Broken is meant to say it doesn't work.

A very very good explaination of the matter
A very very very very good deeper math based explaination
Author's implementation code

Oh and something for market efforts on the same method

Does Allegskill have a Mission Statement?

MrC is right - tau, one of the parameters, is what causes sigma to go up. This is added to sigma (or rather, variance which is sigma^2) BEFORE the new sigma is calculated.

The key concept here is the sigma will go down MORE if the outcome was more surprising and LESS otherwise. So, if tau is sufficiently large, it can override the change in sigma.

I think the first thing to experiment with is a larger tau value.

I should also add, as it is, early convergence for a newbie is a real problem and it affects vets too. From personal experience, when I started playing again last year, I already had some 30+ losses in the database and a handful of wins, potentially from random hider late night sessions. Afterwards my rank fairly rapidly converged to 8. Problem is the delta is now so low, I'm not going to get anywhere near the rank I would have if I didn't have those 30+ losses early on. This is the exact same situation newbies may find themselves in.

The reverse could happen too, one could find oneself with an artificially inflated rank due to lucky early wins.

I think the problem here is, all Alleg games are not equal. We expect those variances to be irrelevant over a large sample size. Trueskill assumes all games have equal weight (except for the skill levels obviously), and under that assumption, the model works.

But if you are in a number of games early on where rank difference is not indicative of the probable outcome (rock distribution, bad commanders, bad settings, bad map for one faction etc) then your rank can rapidly converge to a value that's not indicative of your true skill.

That's why a sliding window will help and could change your ranking significantly.

why not just reset ranks with hider resets?