Tiers in profile

So this is a feature people use very little or at least admit very little using, but I am a stats faggot, look at my achievements regularly and check my tiers in profile, so I feel like this is worth discussing

I feel like there is some kind of imbalance in tier profile, most people in duel tend to fall it tier 2 or 3 while getting much higher ranking in ca. I consider myself as a duel-focused player, and have tier 2 duel, vs tier 7 ca which I always considered myself being average-ish and quite bad compared to top caers which outdamage me by a 2 for 1 ratio.

More annoying, it look a lot win-ratio based since you can see players getting optimized tier while being completly comparable to inferior tier players. Example here, crizis and falconer are about the same level I think, at least there should never be more than 2 tiers diff but crizis is tier 7 and falky tier 2 after more than 100 games each which is completly unfair to their respective levels.

The causes lay in two facts : first, iirc tiers are calculated by frags, so if you stomp on an opponent it may be low ranked each frag still matters for you own elo and so stompers get better tier in duel while they really shouldn't. There is a point where you know you own an opponent and can get as many frags as you want and you should get absolutely 0 reward for them, while in elo you can still get some and progress in ranking stomping on lower ranked players. Second, I think the tiers may be miscalculated, and rely on elo differences instead of relying on populations. The top 10% should be tier 10, 10 to 20 tier 9 and so on ; and not 1 tier every x elo points which create pyramidal system where there is very little people on top tier and a lot on tier 1 and 2.

So this is some thoughts on what should be done but obviously i am not your boss, i don't know better than you this is just my opinion and you, dev team, are free to ignore it completly.
(updated 2012-09-05 17:11:02)

The causes lay in two facts : first, iirc tiers are calculated by frags

Nope.

Ok

edit : Other disturbing factor
<crizis> playing vs unlogged players = gets rated vs default starting value

But some people are very high skill and play unlogged, thus they are under-rated and are punishing to loose to, so these games should be less significant than ranked vs ranked games (for ex. half less lost and gained points)
(updated 2012-09-05 17:22:19)

BBQsauce wrote:

But some people are very high skill and play unlogged, thus they are under-rated and are punishing to loose to, so these games should be less significant than ranked vs ranked games (for ex. half less lost and gained points)

Playing against unlogged players shouldn't count at all.

improving these stats would be awesome! an elo-like rating system would probably be best? in the end it should not be about frags/dmg in duel, but who you win/lose against.

jannis wrote:
improving these stats would be awesome! an elo-like rating system would probably be best? in the end it should not be about frags/dmg in duel, but who you win/lose against.

It already is elo-like in fact
(updated 2012-09-05 17:50:03)

improving these stats would be awesome! an elo-like rating system would probably be best? in the end it should not be about frags/dmg in duel, but who you win/lose against.

That's exactly how it works.

Maybe toukka could give deeper explanation of the rating algoritm.

Hi all !
As a novice in all this stuff I tought I could give a little opinion ^^

The way we are ranked is still very confusing to me, I didn't read that much things about it but I actually don't understand what tier corresponds to which level of skill. I mean, obviously, Tiers 1 players are supposed to be "less skilled" than some Tiers 9 players, but what does all this change in game ? Are we placed in a team by the server based on our Tier or something, cause it doesn't seems to work this way :/

Maybe something clearer to read and to understand how it works would be great, at least for players like me. Just sayin'.

Crizis, you say that it's actually an elo-like ranking system that's in game. But where can I see it ? I guess it's linked with the "Tiers" stuff, but why don't we make a visible ranking like maybe SC2's one, or LoL's one ?

I know it's easy to say, and maybe not really easy to do, but it's just an idea from a normal player ^^

PS : Sorry for crappy English ^^

Crizis, you say that it's actually an elo-like ranking system that's in game. But where can I see it ?

Tiers in your profile is the only place which uses skillranking for anything right now.

BBQsauce wrote:
I feel like there is some kind of imbalance in tier profile, most people in duel tend to fall it tier 2 or 3 while getting much higher ranking in ca.

Thats probably because relative to the player base of that gametype, you simply are better in CA. Or you constantly play against better opponents in duel and you loose a lot.

BBQsauce wrote:
More annoying, it look a lot win-ratio based since you can see players getting optimized tier while being completly comparable to inferior tier players. Example here, crizis and falconer are about the same level I think, at least there should never be more than 2 tiers diff but crizis is tier 7 and falky tier 2 after more than 100 games each which is completly unfair to their respective levels.

The rating system boils down to win/loss in duel because your performance is factored by scores/second relative to other players in that game. In duel there is only 2 players so you either score more or less than the other guy.

BBQsauce wrote:
But some people are very high skill and play unlogged, thus they are under-rated and are punishing to loose to, so these games should be less significant than ranked vs ranked games (for ex. half less lost and gained points)

The amount of change in rating depends on a) opponents relative rating against yours b) internal value that describes the reliability of your rating which increases the more games you play and closer to the estimated result your result in a game is and decreases in opposing factors as well as how long it was since you played your last game.

BBQsauce wrote:
The top 10% should be tier 10, 10 to 20 tier 9 and so on ; and not 1 tier every x elo points which create pyramidal system where there is very little people on top tier and a lot on tier 1 and 2.

The rating system itself works so that it creates a natural gaussian distribution of player skills. So that major population will be tiered around the center tiers.

Warry wrote:
The way we are ranked is still very confusing to me, I didn't read that much things about it but I actually don't understand what tier corresponds to which level of skill. I mean, obviously, Tiers 1 players are supposed to be "less skilled" than some Tiers 9 players, but what does all this change in game ? Are we placed in a team by the server based on our Tier or something, cause it doesn't seems to work this way :/

Behind the tiers is an floating rating system similar to Elo or Trueskill. So behind the curtains you are assigned 2 values which are the floating rating value and uncertainity factor. Floating rating value is a raw value that has no upper or lower limits per se, only limits are in how much that rating will change after each game. I believe I posted this fact earlier when we announced the tiers, but recall that your tier is basically decimated value of your rating relative to the minimum and maximum rating values currently in our dataset in that specific gametype.
So roughly your tier is floor((rating - minrating )/ (maxrating - minrating + 1) * 10) + 1.

Well I think I already had most of that insight already

toukkapoukka wrote:
Thats probably because relative to the player base of that gametype, you simply are better in CA. Or you constantly play against better opponents in duel and you loose a lot.

You are pretending it's just me ? Well this is true in my case, i do play better opponents, i play them logged and unlogged and this is probably ruining my elo and the reason of my poor ranking on that gametype. Unlike you, however, I am uncomfortable about the fact progressing in a ranking system is about farming underskilled matches especially considering we have skill gap issues and difficulties to find skillmatched games for begginers.

toukkapoukka wrote:
The rating system itself works so that it creates a natural gaussian distribution of player skills. So that major population will be tiered around the center tiers.

This theory only is true if there is no elo gained from unraked matches, no leaving of bad ranked players, also there is obviously fluctuation induced from small sample, not-so-random opponent choosing and playing frequency behavior. So, I would bet quite confidently the player distribution in tiers isn't a natural gaussian around center tiers.

toukkapoukka wrote:
So roughly your tier is floor((rating - minrating )/ (maxrating - minrating + 1) * 10) + 1.

There is a lot to be said about this, even theory-wise, taking median for average or min and max difference as a mesurement of standard deviation is all wrong assumptions, these as they are in fact likely to differ, min and max rating each depending of a single person's rating you can possibly have high variance in middle rating tiers just determined by how well or bad do the extremums playing their matches.

Also, I wanted to thank you for explaing even tho it turned out I didn't need more explanations. I didn't want to sound agressive but mostly I got the feeling you are answering questions I wasn't asking for and on the contrary not answering issues I point out, like this completly unrelated q/a.

BBQsauce wrote:

But some people are very high skill and play unlogged, thus they are under-rated and are punishing to loose to, so these games should be less significant than ranked vs ranked games (for ex. half less lost and gained points)

toukkapoukka wrote:
The amount of change in rating depends on a) opponents relative rating against yours b) internal value that describes the reliability of your rating which increases the more games you play and closer to the estimated result your result in a game is and decreases in opposing factors as well as how long it was since you played your last game.

(updated 2012-09-06 12:29:27)

hugteddy wrote:

BBQsauce wrote:

But some people are very high skill and play unlogged, thus they are under-rated and are punishing to loose to, so these games should be less significant than ranked vs ranked games (for ex. half less lost and gained points)

Playing against unlogged players shouldn't count at all.

tl;dr, but imho servers should be either ranked or unranked, so that this wouldnt be an issue.

Well let's agree that currently, the tier thing seems to work strangely. If tier 10 is the highest tier, and tier 1 is the lowest tier, why are many decent players in tier 2/3?

Not that I'm very strong at the moment, but I'm in tier 2 now. I would expect that new players are in tier 1-4, normal players in tier 5-7 and the best players in 8-10 or so. Then the higher your tier, the more difficult to get promoted, so tier 10 is only for a few top players.
(updated 2012-09-06 16:45:38)

BBQsauce wrote:
I didn't want to sound agressive but mostly I got the feeling you are answering questions I wasn't asking for and on the contrary not answering issues I point out, like this completly unrelated q/a.

I was responsing to your statement by saying that the rating of the opponent relative to yours is not the only factor in play here. If you have played multitude of games recently then the amount your rating changes gets smaller.

jannis wrote:
Well let's agree that currently, the tier thing seems to work strangely. If tier 10 is the highest tier, and tier 1 is the lowest tier, why are many decent players in tier 2/3?

We have studied the ratings in our database and it appears that at least duel and ffa have the top of the distribution curve much lower than the average rating.
There might be natural explanations for this. One being that new players come in to play a few games ending up losing them and having their ratings dropped down and then they just stop. We didn't utilize the amount of games player played or when they last played in our study.
Another factor in the other end is that the better players are that much better that their rating keeps inflating as time goes on which seemingly unbalances the system.

toukkapoukka wrote:

We have studied the ratings in our database and it appears that at least duel and ffa have the top of the distribution curve much lower than the average rating.
There might be natural explanations for this. One being that new players come in to play a few games ending up losing them and having their ratings dropped down and then they just stop. We didn't utilize the amount of games player played or when they last played in our study.
Another factor in the other end is that the better players are that much better that their rating keeps inflating as time goes on which seemingly unbalances the system.

Hmm, my observations are:
1. Mostly play players a bit weaker than you (and win) -> high tier
2. Mostly play players stronger than you (and lose) -> low tier

Whereas it should not matter that much.

But it seems that you have a pretty fancy algorithm set up, so it should be possible to tweak some settings I guess?