AllegSkill Online

[badpazzword]

MrChaos is also correct.

[Correct]

Not exactly, he seems to be trying to contradict you and failing.

[badpazzword]

I thought he was more like "errm wtf I don't know what the hell you're speaking about stop making my head ouchy with those unimportant maths here's what I know for sure."

[Vlymoxyd]

Rereadeading the *player skill* and the replies, along with some more thoughs made me find what I think is the answear to my question, Thanks
Since AS just seem to add the "skills" of every player to give a skill to a team, it must somehow mean that the skill is an absolute number(Otherwise, adding up would make no sense).
Also, my logic about thinking that the MU might be a relative skill measure was a bit flawed and giving it more though fixed it up.
Now, I know that balancing using ranks can make sense.

BTW, I wasn't talking sigma underestimating the rank of 99% of the players, this is a completely different problem.



But rethinking about the sigma problem made me think about something:
After a week or so with AS, I must say it's an improvement over helo, but there's still problems when ranks are used to balance teams and find out stacks.

I think the main problem is that the current formula behind rank isn't built to help comms/players to balance teams, it is built to be sure of not overrating anyone.
I think that for a leaderboard(And probably "free-for-all" games with lots of 12 years old), the conservative rank (MU- 3 sigma) is probably the best way to go. However, for the in-game rank, it is very weak at helping people to balance games(And everyone knows why).
Would it be possible(I mean easily) to have separated ranking for the leaderboard and in-game ranks?



So an idea:
Keep the same formula for the leaderboard

For the in-game rank:
-I think that for vets, there should be no Sigma modifer. The MU can be either overrated or underrated. When trying to balance a game using ranks, an underrated rank is just as bad as an overrated one.

-I beleive that the rank of new players should have a bias that lowers it(Because we all know the average skill level of a new player is usually lower than the average player).

So quick suggestions:


In-game rank = MU - (T x Sigma)
T would be a value that starts at 3 and goes down with the time played by the player.


Another idea:
Someone suggested to go for a 95% confidence rather than a 99% one. You'd have newbies starting as (5)s. A (5) would probably be closer to the real skill of a newbie than a (0). The rank of everyone would go up, so a (5) would also be be farther from the "average" than it is currently.


Any ranking formula will have flaws. Choosing one is more a matter of taste than logic, but imo, the most usefull in-game ranking for Allegiance would be one that is the most effective at evaluating the skill of a team rather than one that is best for a leaderboard.

[Tigereye]

w00t leaderboard has been tweaked for more performance.

Who notices a speed increase? (with a current browser. Old browsers still load at the old speed)

--TE

[madpeople]

QUOTE (Correct @ Jan 26 2009, 09:44 PM) Not exactly, he seems to be trying to contradict you and failing.

actually, they are kind the same thing

it is the peak probability that you have a certain rank, and that rank is comparable to everyone else...

[sgt_baker]

QUOTE (Vlymoxyd @ Jan 27 2009, 01:34 AM) Rereadeading the *player skill* and the replies, along with some more thoughs made me find what I think is the answear to my question, Thanks
Since AS just seem to add the "skills" of every player to give a skill to a team, it must somehow mean that the skill is an absolute number(Otherwise, adding up would make no sense).

On the contrary, AS doesn't 'simply add up the ranks' (ranks being the absolute sumber). Please re-read the player rating section of the wiki. You're looking for Mu and Sigma in the 'Forming the Teams' section. Player skill never has been, and never will be an absolute number. They are always expressed in terms of mean and standard deviation, which by their very definition, defines a range is possible skills.

QUOTE Also, my logic about thinking that the MU might be a relative skill measure was a bit flawed and giving it more though fixed it up.
Now, I know that balancing using ranks can make sense.

BTW, I wasn't talking sigma underestimating the rank of 99% of the players, this is a completely different problem.



But rethinking about the sigma problem made me think about something:
After a week or so with AS, I must say it's an improvement over helo, but there's still problems when ranks are used to balance teams and find out stacks.

I think the main problem is that the current formula behind rank isn't built to help comms/players to balance teams, it is built to be sure of not overrating anyone.
I think that for a leaderboard(And probably "free-for-all" games with lots of 12 years old), the conservative rank (MU- 3 sigma) is probably the best way to go. However, for the in-game rank, it is very weak at helping people to balance games(And everyone knows why).
Would it be possible(I mean easily) to have separated ranking for the leaderboard and in-game ranks?



So an idea:
Keep the same formula for the leaderboard

For the in-game rank:
-I think that for vets, there should be no Sigma modifer. The MU can be either overrated or underrated. When trying to balance a game using ranks, an underrated rank is just as bad as an overrated one.

-I beleive that the rank of new players should have a bias that lowers it(Because we all know the average skill level of a new player is usually lower than the average player).

So quick suggestions:


In-game rank = MU - (T x Sigma)
T would be a value that starts at 3 and goes down with the time played by the player.[/quote]

The forumla is defined as Rank = mu - k x sigma. k being the 'k-factor'.

You're forgetting that sigma tends towards ~0.8, which replicates the effect you're trying to produce by manipulating k. k can be whatever value we choose, but must remain constant for all players.

Take Sheff an an example of a stabilised player. Rank = 37.01 - 3 * 0.79 = 34.64. Is the 99% conservative margin really underestimating his skill?

QUOTE Another idea:
Someone suggested to go for a 95% confidence rather than a 99% one. You'd have newbies starting as (5)s. A (5) would probably be closer to the real skill of a newbie than a (0). The rank of everyone would go up, so a (5) would also be be farther from the "average" than it is currently.[/quote]

Yes. That was my idea.

QUOTE Any ranking formula will have flaws. Choosing one is more a matter of taste than logic, but imo, the most usefull in-game ranking for Allegiance would be one that is the most effective at evaluating the skill of a team rather than one that is best for a leaderboard.[/quote]

I've already mentioned that the method used for calculating and displaying team skills in-game needs updating. I can't get everything done at once. The Alleg code is an unruly beast, and this will take time. Give me a chance here.


B

[Raveen]

Ok, here's an idea to slightly improve the rank distribution (there's a lot of Inters and very few vet or aboves). Currently you're doing:

Conservative Rank (CR) = Mu -3xSigma

Rank = CR x 30/50

Now as Sigma cannot go lower than 0.8 the real top CR is not 50 but (50-(3 x 0.8)) = 47.6

ie. Nobody can get a CR higher than 47.6.

So the Rank should be equal to CR x 30/47.6

Of course, monkeying further with this figure could give a better distribution but would be less mathematically justifiable.

Addendum: I suppose you could go for Rank = (30/top CR in database) which would work as long as you don't remove inactive folks from the calculation.

[finki]

For rank distribution:
Why not simply round up after .5? (so you see rank 15 in game if he/she has 14.5 for example)
That doesn't solve your problem that much, but is an easy first fix

It also feels that rank8 is a bit too high to be the minimum threshold for weak players with much playtime. There are so many 8s nowadays. Maybe make it 4 (newbie server protection wise, so they cant in any case play on newbie server. if newbieserver limits lets you play with 4 but not with 5 anymore make it 5)

[sgt_baker]

QUOTE (finki @ Jan 27 2009, 02:35 PM) It also feels that rank8 is a bit too high to be the minimum threshold for weak players with much playtime. There are so many 8s nowadays. Maybe make it 4 (newbie server protection wise, so they cant in any case play on newbie server. if newbieserver limits lets you play with 4 but not with 5 anymore make it 5)

I agree. Will be sorted.