Player Ranking in Allegiance

Naboki · Post by **Naboki** » Fri Dec 21, 2007 1:04 pm

So i now readed the page about True skill, and thats a mathematical not corrcet system.

The standard uncertainy in gamers skill counts only 2! This is wrong when you want an Gauss curve seperation from players. It has to be 3
but yeah take this wiki page too see what i mean.
http://en.wikipedia.org/wiki/Statistics

the first picture exactly tells what i mean ->with the true skill algorithm, there will be 4,6% players will not fit in the curve.

and elo and helo is proven, and they work

For example, the whole chess world decide on elo ratings about grandmasters and so on...
there are elo ratings in soccer and so on....

But thats statistic. Statistic is a science of probatilities.... so there are Failures... thats nature....

sgt_baker · Post by **sgt_baker** » Fri Dec 21, 2007 1:10 pm

Naboki wrote:QUOTE (Naboki @ Dec 21 2007, 01:04 PM) So i now readed the page about True skill, and thats a mathematical not corrcet system.

The standard uncertainy in gamers skill counts only 2! This is wrong when you want an Gauss curve seperation from players. It has to be 3
but yeah take this wiki page too see what i mean.
http://en.wikipedia.org/wiki/Statistics

the first picture exactly tells what i mean ->with the true skill algorithm, there will be 4,6% players will not fit in the curve.

and elo and helo is proven, and they work

For example, the whole chess world decide on elo ratings about grandmasters and so on...
there are elo ratings in soccer and so on....
But thats statistic. Statistic is a science of probatilities.... so there are Failures... thats nature....

In which case you didn't read it properly. /wink.gif" style="vertical-align:middle" emoid=";)" border="0" alt="wink.gif" /> You might wanna try reading ftp://ftp.research.microsoft.com/pub/tr/TR-2006-80.pdf before declaring the system invalid.

sgt_baker · Post by **sgt_baker** » Fri Dec 21, 2007 1:13 pm

The MS site is intended to provide the layperson with an overview of TS. Only the linked paper contains the full functioning of the system.

P.S. To calculate the possible rankings for a player with 99% confidence one uses the formula R = Mu +- 3 * Sigma. Three standard deviations, right?

sgt_baker · Post by **sgt_baker** » Fri Dec 21, 2007 1:27 pm

Naboki wrote:QUOTE (Naboki @ Dec 21 2007, 01:04 PM) and elo and helo is proven, and they work

Truskill actually reduces to an Elo equation when both teams are comprised of players with identical Mus and Sigmas. (Which is also stated somewhere on that site) Elo is proven for situations where single entities are rated vs other single entities (1 vs 1, like chess), but not for situations where teams are involved.

Dark_Stalion · Post by **Dark_Stalion** » Fri Dec 21, 2007 2:52 pm

Purposes of the ranking system in allegiance:

1. To seperate people and allow commanders to get a general idea on who to assign to certain tasks. Leaving the newbies to fly around by veterans and learn the ropes.

2. To allow newbies to practice individually on their own servers, blocking out high ranks who think whoring them in funny.

3. To allow #'s a little elbow room in the boot department.

4. To generally know skill levels and how long players have been here.

5. To tell who can command (which doesn't always work because some newbies and novices have commanded really well in the past).

Do I think anything needs to be changed? I don't have the notion in my noggin about it.

Dark_Stalion · Post by **Dark_Stalion** » Fri Dec 21, 2007 2:53 pm

Thanks for the ranking system, but truly I can't criticize on it. After all I don't pay for it, maintain it and such...

Naboki · Post by **Naboki** » Fri Dec 21, 2007 2:59 pm

sgt_baker wrote:QUOTE (sgt_baker @ Dec 21 2007, 02:27 PM) Truskill actually reduces to an Elo equation when both teams are comprised of players with identical Mus and Sigmas. (Which is also stated somewhere on that site) Elo is proven for situations where single entities are rated vs other single entities (1 vs 1, like chess), but not for situations where teams are involved.

http://www.eloratings.net/europe.html

doesnt look to wrong the elo rating or?
Italy is worldmaster in soccer, france vice....

And from the link of the deep inside Trueskill, i cant see that they use a 3 standard deviations, i onlysee 2

takingarms1 · Post by **takingarms1** » Fri Dec 21, 2007 3:19 pm

I believe the difference is that those ratings are rating teams, not individual players. TrueSkill attempts to rate the players, not the teams, since rating teams is useless to us.

For instance, while ELO might be useful to rate squads, it would not be very useful to rate each player on the squad.

Post by **Raveen** » Fri Dec 21, 2007 3:19 pm

That's fine for ranking teams but it's not ranking players within a team unless I missed something. ELO ranking of teams would be fine because it's a series of 1v1 encounters, much like chess.

You could use ELO to rank each squad based on SG performances but not all players in each squad are equal.

sgt_baker · Post by **sgt_baker** » Fri Dec 21, 2007 3:23 pm

Naboki wrote:QUOTE (Naboki @ Dec 21 2007, 02:59 PM) http://www.eloratings.net/europe.html

doesnt look to wrong the elo rating or?
Italy is worldmaster in soccer, france vice....

They're treating the teams as individual entities, not as collections of players each with their own Elo rating.

QUOTE And from the link of the deep inside Trueskill, i cant see that they use a 3 standard deviations, i onlysee 2[/quote]

Please explain exactly where you're getting this figure of 2 std devs from. I'm not entirely sure I follow what you're getting at.