AllegSkill Online

[sgt_baker]

QUOTE (ogorass @ Jan 19 2009, 12:21 PM) One more thing: do I get it right, that negative stacker rating means 'antistacker'?

That's roughly what it means, yes.

[sgt_baker]

QUOTE (finki @ Jan 19 2009, 11:36 AM) When will commander ranks be shown in Alleg?

There is still some work prending to upgrade the Alleg server and client to fully support the new system. Enhanced in-game stats, such as commander ranks, and a new autobalance system form part of this effort.

B

[sgt_baker]

QUOTE (CrazyFingers @ Jan 19 2009, 03:52 AM) I just read and reread it. It said that the sigma starts at 25/3 and stabilizes at 1 with 1 having the most certainty. But how can you have a sigma lower than 1 if it stabilizes there? Uber reliability?

QUOTE (AllegSkill Wiki)Note that players stabilise with a σ of approximately 1.[/quote]

0.8 is approximately 1. The lower your sigma the more certain the system is of your mu. Sigma could also be held as being inversely proportional to experience.

[Zruty]

BTW, does Sigma grow over time, when the pilot is inactive? Or, in other words, does AllegSkill count all games ever with equal coefficients? Or do the 'late' games contribute less to the rating?

[sgt_baker]

QUOTE (Zruty @ Jan 19 2009, 01:52 PM) BTW, does Sigma grow over time, when the pilot is inactive?

No. Both mu and sigma remain static whilst a player is inactive.

QUOTE Or, in other words, does AllegSkill count all games ever with equal coefficients? Or do the 'late' games contribute less to the rating?[/quote]

The magnitude of sigma dictates how much your mu can change after any given game. High simas allow for large changes to mu, wheras once the system is confident in your rating (low sigma) your mu will change slowly.

B

[Zruty]

QUOTE (sgt_baker @ Jan 19 2009, 05:06 PM) The magnitude of sigma dictates how much your mu can change after any given game. High simas allow for large changes to mu, wheras once the system is confident in your rating (low sigma) your mu will change slowly.

Please, I've read the wiki, and, being a mathematician myself, I pretty much understand the theory behind this.
My question was about rating increase. I mean, a real skill increase, like in my case: I know more and more each time I play, so my AllegSkill rating should increase. But my older games, when I was young and dumb, keep dragging me down.

This means that if I created a new account and played all the time the same way I do now, my AllegSkill would be higher. Sometimes, much higher.

I've met a similar project a while ago (anyway, the ideas behind AllegSkill and TrueSkill are not that new). There, if a user wasn't playing, his Sigma slowly increased, so that his later successes and failures would have more impact on Mu (and on the rating).

[mcwarren4]

Something the leaderboard needs to show is the ingame ranks. What is Weedman's rank? Is it Expert 9?

[Clay_Pigeon]

It is right there under rank. 23ish is Expert 2.

[HSharp]

QUOTE (mcwarren4 @ Jan 19 2009, 03:10 PM) Something the leaderboard needs to show is the ingame ranks. What is Weedman's rank? Is it Expert 9?

The bit where it says rank, so weed is rank 23, 23/7 = 3.28 so is in expert range as 1>=novice 2>=inter 3>=vet and 3<Expert, 23 mod 7 = 2 so weed is Expert 2

[sgt_baker]

QUOTE (Zruty @ Jan 19 2009, 03:06 PM) Please, I've read the wiki, and, being a mathematician myself, I pretty much understand the theory behind this.
My question was about rating increase. I mean, a real skill increase, like in my case: I know more and more each time I play, so my AllegSkill rating should increase. But my older games, when I was young and dumb, keep dragging me down.

This means that if I created a new account and played all the time the same way I do now, my AllegSkill would be higher. Sometimes, much higher.

I've met a similar project a while ago (anyway, the ideas behind AllegSkill and TrueSkill are not that new). There, if a user wasn't playing, his Sigma slowly increased, so that his later successes and failures would have more impact on Mu (and on the rating).

In the case of trueskill, a player's sigma is increased by a small constant value (25/300) every game (dynamics constant - gamma). This prevents stagnation of the rating and allows the system to adjust for a genuine change in skill. Remember that it takes approximately 2500 hours of game play, on average, for sigmas to stabilise. It is also worth nothing that the same rules apply to everyone. In an ideal world we'd have game and account data for every game since the release of allegiance in ~1999. This has obviously not been possible as a result of the game's turbulent history.

The three years of data we do have covers the vast majority of players in this community and gives a pretty good indication of skill. What AllegSkill boils down to is an ability to correctly predict game outcomes 89% of the time purely by considering the skills of either team. This ignores commanders, settings, map, game mode and everything else.

There is also a minor flaw in your logic. If sigma were to slowly increase, shouldn't mu slowly tend toward 25 during periods of inactivity? This is not to suggest that I support the idea for a moment, by the way. More of an indication that there are deeper problems to address when artificially manipulating skill ratings.



Edit: Please READ the wiki next time.

QUOTE (AllegSkill Wiki) is the dynamics variable, which prevents sigma from ever reaching zero, which in turn determines how quickly mu can in/decrease once sigma has stabilised. If we discover that sigma-stabilised ratings are moving too slowly to reflect genuine changes in skill, we will increase gamma.[/quote]

Additionally, I see no problem with the concept that changes in skill are inextricably tied to games played, not time spent idle. If anything, idle time might lead to a genuine, albeit temporary, decrease in skill.