Free Allegiance Forums

You guys don't have the frame that keeps the headers visible? I'm using FF3 and I can scroll and it works fine? has been for 2-3 days if you haven't looked lately, look again.

SaiSoma wrote:QUOTE (SaiSoma @ Jan 24 2009, 04:34 PM) You guys don't have the frame that keeps the headers visible? I'm using FF3 and I can scroll and it works fine? has been for 2-3 days if you haven't looked lately, look again.

Google Chrome says no.

Vlymoxyd wrote:QUOTE (Vlymoxyd @ Jan 24 2009, 01:51 AM) Baker, I took the time to edit my post in order to add that I wasn't questionning the system(Or the maths)... I wish you wouldn't have stopped reading before it...

Yeah I was being abrupt in any case... which I completely failed to communicate in my edit. Long night at the pub n'all that.

QUOTE Manually balancing using ranks or just being able to tell if there's a stack just because a team has higher total rank is something I'm not sure makes sense right now and I'd like to know if it does. I just figured asking the guy would be faster than reading the maths for the 10ths time and giving it a really hard though trying to figure out if what I was thinking made sense. I've taken stats both in college and university, I've read the big MSR article on trueskill and that wiki article but I still lack the knowledges required to fully understand what the MU truly means(Beside the obvious "Higher" = better).[/quote]

Well for a start you're about 1000x more qualified than I am in the field of stats!

Defining mu in 'three simple words' is actually quite a difficult task, and there are various attempts and explanations attempting to do so floating around teh interwebs. Probably one of the easiest to grasp is that mu is proportional to you average performance over all games played, taken, of course, with sigma which denotes the uncertainty around said average, against the baseline mu of 25.

The truly WTF factors begin to show their faces when we learn that Trueskill is, in fact, a cleverly disguised online learning system. Effectively, in layman's terms (I don't have 88 years of stats study to put this better), the system looks at the mu/sigma of all players/teams and predicts an outcome. If the real-world outcome differs significantly (correction: at all) from that expected by the first layer of mathematics, the second layer kicks in to 'correct' the mu and sigma of the relevant players so as to reduce the error next time round.

Obviously there are many different methods for dealing with such a process, and it so happens that the chaps at MSR use one which is essentially grounded in statisticy-sounding maths. Numerous other methods exist, amongst them are neural networks, which are vaunted as being able to solve exactly the type of problem MSR have solved using factor graphs (which are spookily similar to neural nets as it happens).

Anyhow, to answer the question: Your mu will rise above 25 if you consistently 'perform' in a manner which is beneficial to your team. The converse holds. Despite it sounding like there is an absolute definition of mu, it is, in my opinion, slightly grey in that respect. I am in communication with MSR with regard to True/AllegSkill, and might pose this very question some day. I'd be interested to see the intellectual logic between mu -> performance -> result.

QUOTE Obviously, a team with a higher total rank should have a tendency to be better than others, but a shorter way to say what I'm wondering:


After taking the hypothesis that the conservative rank = the real skill of all players:
Is the rank an absolute measurement of skill(You can add it up and say that a team is stacked because it has a higher total sum of ranks(Or MU) or just a relative one( Rank (10) + (8) =! (6) + (12) )?


You don't have to answear/reply if you don't feel like it(I won't have any bad feeling about it, really, it's not an important question), but pls don't tell me that ranks aren't used in the formula and that I should read a wiki article that says what MU/sigma means to a player but offers no insight on what it means on a whole team.

I just keep hearing that teams with perfectly equals rank are "balanced", but after giving it a though, I somewhat believe that teams balanced in ranks might not be balanced in skills(Again, even if every one's sigma was 0 and that all MUs were perfectly right).[/quote]

Let's remember that conservative ranks are currently defined as "We are 99% certain that player X is no less skilled than <conrank>. The problem with conservative ranks is that a statistically sounds system, along with all the uncertainties inherent in our task, is forced to produce a single number to define a player's skill. Whilst this very notion frustrates me immensely, I do realise the importance of having a method for communicating the basic facts in a quick and concise manner.

This brings us to the main point: The current system of calculating team ranks in Allegiance is fatally flawed. You were absolutely correct in recognising that there might be problems with adding up lots of individual conservative ranks. I was aware of this flaw before AS's implementation was even a certainty. The correct method for calculating team skills can be found on the 'Player Rating' page of the AS wiki. Once one has arrived at a (capital) mu/sigma for the entire team, one can apply the same conservative rank formula to the team values. It is also worth noting that we can not only denote the current balance (as happens at the moment), but also provide players with a time-sensitive figure which would display team balance with regards to the entire history of that particular game.

I'll say this again: The problems we're currently facing are nothing to do with AS per se (apart from trivial things like stack ratings and base kills/hour), but ones associated with upgrading the game server/clent to fully support the new system and make more sense of the numbers for the user. It is looking increasingly likely that I will have to personally trawl the Alleg code to get the required changes implemented, so this may be a case of Two Weeks™ - or, in the case of AS - Two Years™.

Fortunately, for me (not for you), I have some RL projects I need to concentrate on in the immediate future; I have no eta for getting client/server changes implemented at this time. We all need to earn our bread.

i believe the skill you have according to the system is the weighted area below the graph of the mu-sigma bell using a linear measure that goes from 0 to 1 in [0, 50]. your center is related to mu and your 'bell shape' is related to sigma (as sigma -> 0, it approaches d_mu dm (dirac delta at mu over the measure previously defined).

anyone related to the project care to decrypt my statement and tell me if it is true or not?

*hello* ..@.. discussioning discussioners^^=)

question #1: how about to show up a little js-pop-up when you click onto a field?

question #2: how about to show - when clicked onto field of command wins + losses - a compression of how the density of pilots to the density of win/loss a game?

ps: that could be as well sort of work for our artworkers to generate nice math graphs, don't it?

fufi wrote:QUOTE (fufi @ Jan 26 2009, 12:46 AM) *hello* ..@.. discussioning discussioners^^=)

question #1: how about to show up a little js-pop-up when you click onto a field?

question #2: how about to show - when clicked onto field of command wins + losses - a compression of how the density of pilots to the density of win/loss a game?

ps: that could be as well sort of work for our artworkers to generate nice math graphs, don't it?

why that is a very good idea compacted in a very nice easy to read post, strange.
anyhow, good idea shouldnt be hard to do, thou second one depends on what kind of info on those comm w/l allegskill holds

Evincar wrote:QUOTE (Evincar @ Jan 26 2009, 05:32 AM) i believe the skill you have according to the system is the weighted area below the graph of the mu-sigma bell using a linear measure that goes from 0 to 1 in [0, 50].

No. Mu is simply the peak. Sigma is simply the curve's spread. Rank's just mu - 3 sigma.

The point about dirac delta seems to be correct (as long as you realize that sigma can NOT go to zero).

sgt_baker wrote:QUOTE (sgt_baker @ Jan 25 2009, 06:22 PM) This brings us to the main point: The current system of calculating team ranks in Allegiance is fatally flawed. You were absolutely correct in recognising that there might be problems with adding up lots of individual conservative ranks. I was aware of this flaw before AS's implementation was even a certainty. The correct method for calculating team skills can be found on the 'Player Rating' page of the AS wiki. Once one has arrived at a (capital) mu/sigma for the entire team, one can apply the same conservative rank formula to the team values. It is also worth noting that we can not only denote the current balance (as happens at the moment), but also provide players with a time-sensitive figure which would display team balance with regards to the entire history of that particular game.

I'll say this again: The problems we're currently facing are nothing to do with AS per se (apart from trivial things like stack ratings and base kills/hour), but ones associated with upgrading the game server/clent to fully support the new system and make more sense of the numbers for the user. It is looking increasingly likely that I will have to personally trawl the Alleg code to get the required changes implemented, so this may be a case of Two Weeks™ - or, in the case of AS - Two Years™.

While talking about things to add in game and team ranks, you may want to display the team ranks for each team (similar to as we do now, but using hte new calculation), but also locally (in the alleg client) show what the team's rank would be if the current player were to join that team (show it in brackets).

The online learning thing is a rather nifty idea too, I hadn't thought about trying to apply that kind of idea to this kind of problem.
It kinda reverses the definition of rank, in that your rank is therank you should have had such that the system correctly predicts the outcome of the previous game you played.
Although the outcome it's self is a functon of the team rather than the individual player's rank, over time it should hone in on something near the correct rank, though it will always be one step (game) in time behind th present.
...
It's rather like driving a car while only looking backwards.

Badp wrote:QUOTE (Badp @ Jan 26 2009, 03:03 AM) No. Mu is simply the peak. Sigma is simply the curve's spread. Rank's just mu - 3 sigma.

The point about dirac delta seems to be correct (as long as you realize that sigma can NOT go to zero).

Errrm Mu is where you as an individual are related to everyone else playing Allegiance, Sigma is the uncertainty of your placement in the population.

Badp is correct.