Free Allegiance Forums

Vlymoxyd wrote:QUOTE (Vlymoxyd @ Jan 21 2009, 06:54 AM) I'm just wondering about the stackrating:
I thnik that using the total imbalance can be misleading when you compare players who don't play with similar game sizes.

I concede that the stack rating probably isn't perfect yet. You've made a very good point there. There is already one fix pending which corrects a problem in the fourth dimension (harr! I just wanted to say 'fourth dimension'!), and normalising for team size is definitely something we could implement. Unfortunately the latter fix requires a recalc over the entire database, so will probably have to wait in queue until we have a number of good reasons to perform a recalc. (in order to keep ASGS running smoothly, a recalc is throttled so that it takes appx. 24 hours)

B

RHINO_Mk_II wrote:QUOTE (RHINO_Mk_II @ Jan 21 2009, 11:58 AM) IQ is more flawed than Allegskill!
Good job for getting this online, my rank dropped but that's due to my high sigma, since I haven't been here that long. Many players' ranks (especially squadded players) have dropped quite a bit. Is this because they are more likely to anti-stack, or to probe and nan rather than dogfighting? Please comment.

Very roughly speaking, mu>25 is the likelihood that your contribution will assist with leading your team to victory regardless of which role you're performing in-game, on average. The great thing about the maths behind AS is that it doesn't matter what you do to help your team win (as opposed to a system based on kills, bombing or probing for example), AS will pick up on it. The same obviously goes for losing .

The single-number 'rank' is an approximation of skill, and isn't used anywhere in the vital AS rating calculations - it's designed to be easily digestible by human minds. In the case of AS, 'rank' is stating 'we are 99% certain the player is no less skilled than x'.

It is well worth learning to interpret mu and sigma so as to be able to more accurately assess a player's skill by eye.

Finally, the reason that lots of players' ranks have dropped/moved/done something new is that AS takes a fundamentally different approach to the population as a whole when compared to Helo. All player mus form a normal distribution (bell curve) centered around 25. To put it another way: Most players are average in skill. Helo had a 'newbie helper function' in place in an attempt to artificially reproduce this effect, but this was very much a 'thought after the fact'.

hotnoob wrote:QUOTE (hotnoob @ Jan 21 2009, 03:26 AM) zruty, u seem to be doing math... not programming...

This statement sums up Mr. 160IQ.

You don't make math work to your programming. You make programming work to the math foo'.

sgt_baker wrote:QUOTE (sgt_baker @ Jan 21 2009, 12:20 PM) Very roughly speaking, mu>25 is the likelihood that your contribution will assist with leading your team to victory regardless of which role you're performing in-game, on average. The great thing about the maths behind AS is that it doesn't matter what you do to help your team win (as opposed to a system based on kills, bombing or probing for example), AS will pick up on it. The same obviously goes for losing .

*I am no mathematician*


I am a little confused on this. Surely an anti-stacker who displays great skill, nans miners, probes for HTTs and kills the occasional miner will lose rank over time - just really slowly. If the anti-stacker wins against the stack he will recoup some of this rank - no? The only consolation is that the anti-stacker has a nice negative stack rating to offset their low rank.

I think my philosophical point is just that AS seems to assume you have 'skill' and that you contributed if the team you are on wins. Likewise it assume you lack 'skill' and contributed to the defeat if your team loses.

Lastly, if I understand your point about the bell-curve correctly (high chance I did not) does this not mean that there will be very few people ranked high and similarly few people ranked low - with the majority held around the average?

Confoozled.

heh... atm players on the team with a high chance of losing, drop from play thus causing a huge end game stack.. its funny.

Z

SouthPaw wrote:QUOTE (SouthPaw @ Jan 21 2009, 12:49 PM) I am a little confused on this. Surely an anti-stacker who displays great skill, nans miners, probes for HTTs and kills the occasional miner will lose rank over time - just really slowly. If the anti-stacker wins against the stack he will recoup some of this rank - no? The only consolation is that the anti-stacker has a nice negative stack rating to offset their low rank.

Well they would lose rank if they were constantly losing games.


QUOTE I think my philosophical point is just that AS seems to assume you have 'skill' and that you contributed if the team you are on wins. Likewise it assume you lack 'skill' and contributed to the defeat if your team loses.[/quote]

Whislt we use the word 'skill', what we're actually dealing with are probabilities and likely outcomes. If everyone understood the underlying maths perfectly, we'd never have to refer to it as 'skill' ever again. Alas, this is highly unlikely, so we're forced to use language and numbers that can be understood by the unwashed masses.

QUOTE Lastly, if I understand your point about the bell-curve correctly (high chance I did not) does this not mean that there will be very few people ranked high and similarly few people ranked low - with the majority held around the average?[/quote]

That's exactly what it means.

Zapper wrote:QUOTE (Zapper @ Jan 21 2009, 12:55 PM) heh... atm players on the team with a high chance of losing, drop from play thus causing a huge end game stack.. its funny.

Z

Really? I don't see what they're trying to achieve. AS doesn't consider the likely outcome based on team balance at the end of the game - it's based on balance over the entire course of the game. Creating a massive instantaneous stack by dropping doesn't significantly reduce the rating reduction applied to players on the losing side, and will only lead the the game ending sooner than would have been the case had they stuck to their guns - forther negating the attempts at 'damage control'.

I'll keep an eye out for this should it become a problem, though.

B

sgt_baker wrote:QUOTE (sgt_baker @ Jan 21 2009, 11:59 AM) I concede that the stack rating probably isn't perfect yet. You've made a very good point there. There is already one fix pending which corrects a problem in the fourth dimension (harr! I just wanted to say 'fourth dimension'!), and normalising for team size is definitely something we could implement. Unfortunately the latter fix requires a recalc over the entire database, so will probably have to wait in queue until we have a number of good reasons to perform a recalc. (in order to keep ASGS running smoothly, a recalc is throttled so that it takes appx. 24 hours)

B

The stack rating is quite imperfect, but you stated as much before. It can often be a measure of how much people stack you as much as you stack them as far as I understand it. This explains why some of the top players have high stack ratings (people see them on a team, and join that team because of it) and also is a theory for why FedMan has such a low stack rating (people see him and join the opposite team).

is there any point in scaling down the ranks by a 1.6 factor? why is the number 25 (30?) so important? this ranking doesn't have anything to do with helo or elo, why should it have numbers that try to represent the same thing? the more precise the numbers, the more balanced games can be (especially small games).

Evincar wrote:QUOTE (Evincar @ Jan 21 2009, 10:57 PM) is there any point in scaling down the ranks by a 1.6 factor? why is the number 25 (30?) so important? this ranking doesn't have anything to do with helo or elo, why should it have numbers that try to represent the same thing? the more precise the numbers, the more balanced games can be (especially small games).

This is because the "Veteran 2, Novice 5" etc etc ranks exists as a text file on every Alleg server. Currently the servers are incapable of dealing with ranks outside of the range 0 - 30. My next targets are the Alleg server and client.

Watch this space.