Free Allegiance Forums

NightRychune wrote:QUOTE (NightRychune @ Dec 25 2010, 01:39 PM) RPS is probably the most balanced core ever for actual gameplay

I used to have a modded version of RPS (I dont remember what I named it) that used DN style treasures (so no floating AB2) and I think i made the GS a bit more expensive since all you really had to do was go Assault Garrison for your tech otherwise.

Sadly I lost it when i reformatted, wouldn't be terribly hard to re-do...

If we could get the new dreg and tf models on it I would play it more.

Add Thorium rocks that build an extra hard outpost. Makes tech availability more unpredictable.

FreeBeer wrote:QUOTE (FreeBeer @ Dec 26 2010, 05:58 PM) Unless I've totally got their intention and implementation wrong. I don't claim to be an expert here. I have not seen the code, nor do I understand the underlying math behind it. I do know that much consideration was given to late joiners and early quitters and I'm confident that they handle this gracefully and with the intent of accurately capturing each player's relative contribution. The goal, after all, is to provide a means to produce balanced games by providing the players and comms with a metric in which to gauge that.

From the mathematical point as will never be accurate ---> as is based on trueskill, trueskill uses the gaussian normal distribution - gaussian distribution you can only use with infinite endings -> Fail, alleg endings from[0,30].

a gaussian distribution implies that you can reach a rank from 999999 or -888888

http://en.wikipedia.org/wiki/Beta_distribution
http://en.wikipedia.org/wiki/Triangular_distribution

would be better

Naboki wrote:QUOTE (Naboki @ Jan 9 2011, 09:11 AM) From the mathematical point as will never be accurate ---> as is based on trueskill, trueskill uses the gaussian normal distribution - gaussian distribution you can only use with infinite endings -> Fail, alleg endings from[0,30].

Fail, µ is not bounded.
Fail, wrong thread.

badp wrote:QUOTE (badp @ Jan 9 2011, 10:34 AM) Fail, µ is not bounded.
Fail, wrong thread.

i (and you should too) dont care about µ, when the definition of a gaussian distribution is already hoaxed

Naboki wrote:QUOTE (Naboki @ Jan 9 2011, 10:51 AM) i (and you should too) dont care about µ, when the definition of a gaussian distribution is already hoaxed

Maybe you need to understand what you're talking about before you critique it; you'll find out the not so meaningful rank number can and does go below 0 (and can go above 30).

i am not here to teach you the basic of a gaussian distribution

it will never fit for any learning process, cause they arent linear (and the skill of players i normaly improveing - so a early played game shouldnt weight less then a recently played game)
mu-sigma*k will never reach the x-axis (and thats what the whole trueskill bla bla is saying) (thats because of the definiton array)

Naboki wrote:QUOTE (Naboki @ Jan 11 2011, 05:51 PM) i am not here to teach you the basic of a gaussian distribution

i am not here to teach you how to use the internet -- but I will.

Please don't a) hotlink images without giving credit and b) don't hotlink or pagelink to sites stealing other's content.

(The original article at least linked to another content stealer who, while still not attributing the source, at least links to it.)

Cort wrote:QUOTE (Cort @ Jan 11 2011, 05:59 PM) i am not here to teach you how to use the internet -- but I will.

Please don't a) hotlink images without giving credit and b) don't hotlink or pagelink to sites stealing other's content.

(The original article at least linked to another content stealer who, while still not attributing the source, at least links to it.)

Who cares. Using the gaussian distrubution is much easier (read: feasible) than using beta or weibull or another invention. I have used the gauss distribution on worse suited problems than ranking allegiance, succesfully.
The gaussian distribution has several advantages.
#1 Many people (read: 1%) understand how it's working.
#2 The MLE which is used for calculation new values after a measurement (even if it's 1-0 result) is a mostly linear calculation if it is guassian. Doing a MLE for other distributions like limited gauss, beta e.t.c. requires the solution of a nonlinear optimization problem which takes into account all games played and uses every players parameters (here:mu,sigma, but they can be different) as variable. Seldom do we know if the algorithm can converge to the correct solution but for gauss is a quadratic problem (read: easy) whose solution is the well known update fomula.
#3 It has worked once when MS used it for trueskill.

For me, as an engineer the argument #3 ranks much higher than anything else.
Allegskill has some unsolvable problems because every game has 1 bit for 20 players (maximum).

Also who cares about stealing pictures, it's the internet after all. I bet 90% of pictures in this forum are stolen and without source. Or more.