my elo is going full retard aswell. its probably 1400 now, but i dont have issues with most 2000+ guys. call me flabbergasted

It's been like that for a long long time.
It's broken and should be scrapped all together!

It has been.

You probably did worse then what the algorithm expected you too, even if you won. That's how I would make my system. It's not there to reward you, it's there to rank you correctly. But I can see why you would be confused. Despite displaying "CA Elo", QLRanks does not use Elo for non-duel gamemodes.

As for the match that you left without playing long and you still lost a lot for it; I think it may be a problem yes.

I don't think i can do worth when i win 10-0 with 30 frags and 3 death man :) and the server was ranked and had autoschuffle based on stats (not randomised shuffle).

edit: but yes i see your point, the system could make a prediction on scores and efficiency of the players but it's not what is does as you can see.

.....

why exactly do you make a secret out of it? is there a nda?

OK OK, I know what you're thinking, "guy plays vs noobs so he has to lose rating".

But without even looking at his profile, I'm going to guess that he's also affected by the "Elo change" problem. Which touches everyone and sometimes you get what looks like -30 +20 jumps in two games all of a sudden.

You know this, as a quick Google search reveals:

January 2012:
http://www.esreality.com/post/2223398/re-qlra...pid2224321
http://www.esreality.com/post/2223398/re-qlra...pid2224373 (yourself admitting "Elo change" has a problem)

April 2012:
http://esreality.com/post/2249210/qlranks-is-...pid2250429
http://esreality.com/post/2249210/qlranks-is-...pid2253495

Normal users don't expect this to happen, so they just take that column and base complaints on it.

i get 5 elo per win!!!!!

Actually, I'll try to understand their system.

Let's take this match from you :
http://www.quakelive.com/#!profile/matches/B4...1fd90/ca/1

Name Score SecondsPlayed EloBefore EloDiff

Your team :
_adrian3k_ 52 783 1033 +2
master_vst 35 778 1381 +0
B4ZE 31 286 1961 -32
Anarkopsykotik -1 8 973 NODIFF (SWITCH)

Other team :
cybille 43 796 1327 +14
manjeriko91 17 477 832 -32
Anarkopsykotik 10 260 +0
Freelancer 2 196 2054 +0 (QUIT)
loveu 11 257 716 +7 (QUIT)

Oh crap. I'd prefer analyzing a match without quits and switches first. That's gonna make things easier. I'll leave that here for ulterior reference and for those who may be interested in trying by themselves too.

Ok fuck this I tried with another match where nobody quit, nobody switched and everybody had the same playtime and I don't understand some of the results. It's probably not working the way I originally thought it would.

In WoW arena if you had a much higher rating than your opponent and won you lost rating. If you are 30/3 it sounds like you're tier slumming. Putting these two together this could explain why you lose rating for your wins.

we have a winner

how can anyone tier slum in ca when tier3 servers are hidden to tier4 players.

On public play, 99,9% of ca activity, that is completely random.

QL's tier system is completely irrelevant here... if doesn't matter if the shitties are Tier 4; they're still shitties.

I know (I throw shit at tier 4 ca all the time). Thing is it is hardly tier slumming if one cannot choose servers to tier slum.

I wonder how i can be tier slumming when server is sv_ranked 1 and i am in tier 4 indeed. It's not like i am doing this on purpose...

edit : i would understand +0 but in some matches there was clearly some 2000 elo players in the opposite teams and i was still rewareded with negative elo :O

All you need for tier 4 is a skillrating > 100 right? So there's still a huge gap between tier 4 and god tier.

i don't know but i don't think so. sv_skillrating goes up to 999. 100 is realy weak i think, concidering it is common to see 999 players like me.

No it's about 70 or 75. Just like ti was before the cap was moved to 999 because pretty much only those stuck at 100 before went on over 100.

Oh that's right cos if you go over 100 it gives you that gold bar underneath your statistics page. But ye the point is, tier 4 has a huge variance of skill.

The caps been 400 for while now, can't remember since when tbh, it may well have been a little before you posted this though still.

It was done to try improve the randomness of shuffles iirc.

Since 3 months after the post you replied to.

does your elo really matter that much? what do you use it for?

There is no efficient way of ensuring that you are not tier slumming by accident. The ELO punishment for a win against a noob team is impropotionate.

It should be a static -1 when you reach whatever the threshold is for a "negative win".

This would ensure constant negative rating for habitual tier slummers without the ridiculous punishment against the accidental match against noobs.

-10 should be the cap imo.

you can check the elo's of players before joining a server and ingame via wn's greasemonkey scripts.

Yup. As I said, no efficient way.

Say your ELO is 1900. Here is the ELO of players on a server:

1830
650
1700
1300
2200
900
1050

Do you join? Will you get negative ELO when you win on that server? Who the fuck knows.

you most likely won't as you will atleast face the 2200 elo guy in the opposing team :)

that's a server i'd join, the one's were everyone's elo is below mine i wouldn't, both because of elo and general uselessness of the players present.

I feel this happens with duel skill rating. when I rape a noob it goes down in rating. If I lose to a good player (but close score line) it will go up :0

who gives a fuck about ELO?

THIS!

I don't really understand how the QLRanks calculation works, but for CA it seems to me that it should work purely on dmg delivered:

Disregard win/loss variable and frags variable (we all know those players who just hide with full stack to pick off easy kills as the last player standing - they get tons of points for this even if their team loses because of it).

Your elo should be calculated purely on the damage ratio you output based on the time you spent in the match, so if you joined the match in the last final minutes, or quit after 10 mins then if your damage output per minute is as high as those who played the entire match then you will be ranked on a par with them.

The weighting of the damage score described above should be weighted up or down based on the average elo of your team vs the opposing team, so that if for example your team's average elo was lower than the average elo of the opposing team and you scored equal dmg with the top scorer of the opposing team then your QLRanks increase should be greater than the top scorer on the opposing team (even if their individual QLRanks level is lower than yours). This reflects the fact that it is more difficult to deal alot of dmg when the rest of your team are noobs...

Don't know if this would work, but the focus on dmg output ratio would simplify things and be a more accurate reflection of individual skill whereas win/loss frags/deaths can be caused by all sorts of extenuating circumstances...

somehting like this. it also could prevent people from not balancing/switching teams if simply the [dmgD / netto-playtime ] is calculated. ignoring wins and frags.

also maybe the maximum value could be capped at 10 pointz +/-. over time this should be more precise than going up or down 100 points for random reasons

I think this is best too. higher skill players will almost always have the most dmg delivered but may not get the kill.

Damage dealt is the most important but frags shouldn't be completely neglected. They are important. A living opponent is a danger. Sure you can decide to let him live after dealing 299 damage instead of chasing and getting hit to frag him but it will likely result in more damage taken for your team as he'll continue doing damage. I think the score in CA is spot-on and it's what should be used.

The QLRanks calculation is not spot on. I have had tons of matches where I finish way above players who are 400 or 500 points higher than me on elo. And similarly players with a TOTAL of 800 or 900 points are often lethal, It's f*cked. If it just kept it simple and stuck to dmg dealt I think it would be a bit more accurate, not perfect like you say a player with 1 hp is still a threat, but whatever tyhe calculation does now it dosn't work, so keep it simple and it might be ok.

The score in CA is simple. Anyway, the difference between score and damage dealt is so small I'd be very surprised if that alone could change QLRanks' outcomes much.

Well right now it's a win/loss thing right? So it's not based on your individual score either.
So even if you play amazingly in every game ever but your team always loses, you will lose ELO.

Maybe the win/loss should weigh in half and your indivudual performance (score or dmg dealt, doesn't really matter as you say) should weigh in the other half. Something like that.

Maybe it already does I dunno...

It's not a win/loss only thing, no. I've gained Elo on losses and lost Elo on wins. Winning or losing seems to have a large influence though. I also can't tell what importance individual performances have. But if it has any importance, the score should be used instead of only the damage dealt in my opinion to have a better and simpler system.

The little system I made quite a while ago on a spreadsheet works exactly like what you suggest. Half for the team and half for the individual. Take a look if you want.

https://dl.dropbox.com/u/74076095/QuakeLive/skillrating.ods

It looks good but there are problems with that. Or there may be; I don't know. I'd need to use it against a population and evaluate the results. The way I see it; it's probably way too "linear". Sure, if everyone was pretty much of the same skill it could work well but it's not like that in a real population. So the problem of transitivity appears. If y does twice better as x, he'll get a rating of 2x. If z does twice better as y, he'll get a rating of 2y. But will z do four times as good as x if they play in the same match? If not, that will screw the ratings up. So the better idea would be to use something more complex than a expectedScore = totalScore*individualSkill/TotalSkill to calculate expectations (on both team and individual sides). There are probably patterns and functions that can be found. Like saying that a team of 300+100+100+100 is stronger than a team of 150+150+150+150 despite the sums being equal. I'm not sure what could be used to help with that. An hidden Markov model, maybe? I'll look into that one day; when I'll be done with other projects of mine.

Agree.

Btw anyone knows how exactly the CA scores ingame are calculated?
It's not dmg(/100) and it's not dmg(/100) + frags

It always was dmg(/100) + frags, wasn't it?

+1 point per 100 damage dealt
+1 point per frag
-1 point for suicide/teamkilling (telefrag on spawn or not)

Follow these steps for easy elo points in CA:
1: "+back"
2: bind x "?????"
3: PROFFIT!

It's fucking boring tho.

well i don't get 1 thing.

for example:

i won a game and got +36 points.
i play another game and get lets say +5
i've noticed, more than once that the 36+ I got earlier changes into a lower amount... whats up with that? :)

I have better net damage and win % of most of the top 20 and I'm not tierslumming since I only join skill-matched public servers as most of the people do, but somehow my rank is shit.

http://www.qlranks.com/ca/player/exp

I don't mind too much about that, but the thing that is bugging me is that as a donor I should have all my matches automatically crawled but yet I always have about 50-60% of my matches not tracked and I don't know why.

Check out my comment #50 comparing your stats to another player.

well i think its clear, CA elo is a load of SHITE. gonna ignore it for the time beeing and get rid of that stupid plugin.

The reason B4ZE's rating looks like it's going down is because of how the rating is calculated:

We have two components, 'rating' and 'accuracy of rating'.

New players start with a rating of 25 and an accuracy of 6.

Accuracy number decreases with more games played, as the accuracy of the rating becomes more.. accurate.

We then calculate the elo figure as following:

40*(RATING - (ACCURACY * K))

This enables us to only put the top players as those with higher accuracy ratings.

When you have surprising wins or losses, the accuracy number increases. This makes your rating change more variable. That enables us to rank players accurately more quickly.

What happened with 'B4ZE' is that his RATING never actually decreased, but the accuracy rating adjusted as the algo was less sure about how accurate the rating was.

Consistent expected performance drops your accuracy rating. Consistent wins increases your rating.

Nobody actually lost 'rating' when winning (unless they are playing total noobs - which is rare). It just seems that way - playing noobs may fluctuate your 'accuracy' rating more.

In short, just play quake.

As an addendum, I may remove the accuracy from the elo figure calculation and just require an accuracy figure of below 2 to appear in the top 100 list.

This would solve a lot of the confusion..

hohum!

sry, was drunk as hell

http://tinyurl.com/6p9so79

Sirax, please consider what people are saying. Let's face it, if you look at the top 10 of each country, you see people in tier 2/3/4 with similar ELOS, even though they are worlds apart skill-wise. This rating system seems flawed. I understand it's because these tier 'ecosystems' don't come into contact with each other, so tier 3 top player never vs tier 4 top player. And therefore can have the same ELO. Maybe the system should take tier into consideration, and adjust elo accordingly.

PERFECT EXAMPLE:

I looked at exp's link to his QLRanks profile, and compared him to no1 in italy.

Exp's stats: World Rank #786, CA ELO 2122, WIN 71%, AV DMG Dealt 5602, AV DMG Taken 3693.
Teuz4's stats: World Rank #8, CA ELO 2579, WIN 60%, AV DMG Dealt 4786, AV DMG Taken 4104.

This is crazy! exp has a much higher win ratio, he delivers FAR more damage and takes FAR less damage. Stats wise he is far superior, and yet the rating system ranks him as a much worse player than Teuz4.

This system is completely flawed.

Your example is useless because we don't know who their opponents are.

Anyone can have super pro-looking stats when playing against noobs. If I play your mom all day every day and take zero damage, and 99% wins, should I be ranked #1?

Also he only has 560 tracked games.


Quick edit: The system may still be fucked. I just don't agree with your example.

Whether I use examples or not, it still is a fact that the top10 list of any country contains players from different tiers. Top 10 should _only_ include top tier players, accordingly ranked above players in lower tiers.

So your perfect example turned out to be shit, but you are still correct because... you say so?

I'm sure you could actually find some weird misplaced player ranked too high. Probably not in top 10s but maybe.

The thing is, it will sort itself out. If they don't "deserve" the high ranking they will drop as they continue playing and losing.
If they continue winning they will remain up there and rightfully so.

You don't understand what I'm trying to say, you also don't understand the problem and you're obviously unaware of the issues with the top 10 ranking of any country.

Oooh let me try that

YOU don't understand what I'M trying to say. You also don't understand that there is no problem and you are ovbiously unaware of the accuracy of the country top 10s.

Now cut the BS and show me some inaccurately ranked top 10 players or stfu.

I'll give you some good examples later.

You just made a really stupid post :(

as rope said, exp has only 560 ranked games and i've never seen him playing premium CA servers. It's easy to have super stats playing random standard servers where you can just +back and rail. Also, if you play good players you must get on certain positions and try to win your 1v1 situation, and not being the 'CA hero' hiding most of the time remaining full while all your teammates are dead.

The system as it stands is completely broken. It should be based on dps while alive/total damage/how you placed compared to the rest of your team.

I disagree

I don't understand the argument here. I dislike discussions of ELO outside of duel as...it's largely silly, irrelevant, impossible to accurately quantify. What I would say, is that winning should never result in a negative ELO return regardless of the rating of the opposing team. I would say it should result in a null ELO change in this situation. Another thing is, QL doesn't use ELO, it uses skillrating in the shuffles. So, if you are going to ELO track CA or other modes, you should attempt to mirror the results of the skillrating system. Despite the fact that it is broken, it is what game balance is weighted against when playing. The user has little control over this unless he decides to abuse the system and and force himself into a favorable situation for ELO advancement. It doesn't make sense to punish a player in an ELO ranking when he is simply balanced into a team based on entirely different criteria from the tracker.

And finally: We are discussing CA ELO... The real answer, stop trying to track CA ELO rankings?

Thanks for posting that. Saved me the effort. Shame I can't plus you (looks like I can! :D)

As I see it you either track CA ELO in a better way or you don't do it at all.

I completely agree.

sirax was told about it multiple times but he likes his way of thinking more than everyone else's. (Damage made)/(time played) with some condition on minimum time played like the -accuracy part is the only applicable solution for CA mode. That is, if having elo in team-based games is applicable at all in the first place.

Does the ranking system take into consideration how good a specific player is on a map in duel?

Example:

Player A has 1500 rating on Toxicity
Player B has 2400 rating on Toxicity

Player A has +200 rating vs Player B on the rest of the maps.

Both players have 2200 ELO.

Player B beats player A on Toxicity five times in a row and gains ELO based on their player ELO instead of a formula consisting of player ELO + map rating value.

Has that been implemented? If not, why not?

The system does also not take into consideration if you just jump into a team with much lower total ELO compared to the team you're facing. Something that I don't mind doing since I'm all up for the challange.

What I don't like is that even though I manage to carry the team alone and end up with some really good stats I still end up losing a lot of ELO from what I've seen. I remember thinking that it looked odd that I had lost 100 ELO one week while I had been playing really good in those few games I can get after work.

You could basically tell your team mates to go afk, put an aimbot on, do 15k damage, lose 10-9 and still end up losing elo or am I wrong here?

I'm positive against qlranks. I think it is fantastic what they've done with it and how it has impacted the activity of the community. What I can't get my head around is how you opt to go for a ranking system that is purely based on wins/losses and some accuracy rating that amplifies the ELO change. Quake is much more than being consistent and winning/losing.

Does quitting in the middle of the game you're losing save you from dropping elo? I've seen some player called ra1den doing that whenever he's about to lose. I've seen a lot of others do it too. Spectators refusing to join the losing team is another one. My point is that the current ranking system directly leads to this sort of behaviour.

You think accuracy/doing damage/other arbitrary figure in a team mode is the most important thing.

I think the team winning is the most important thing.

Funnily enough, my way has come up with a pretty excellent top 10 players for both CA and TDM. Your way would not (I know, i've tried it).

We're just trying to help improve things and think everything you've done is bloody great, but there clearly is room for improvement on the team game ranking system.

I agree that the top 10 players for CA are pretty accurate, but a thermometer that can tell you when water is boiling, but not be able to distinguish between 50deg and 85deg is not very helpful for the majority of cases one wants to use a thermometer. Win/Loss would be the most important variable if you were rankning static teams, but you're ranking individuals based on the win/loss of the team. This is like ranking the quality of a mouse based on the win/losses of the player - performance of the player could be adversely affected by a whole number of other poor PC components when the the mouse itself could be top quality. Conversely, the player could be so good that even though the mouse is complete crap, it is ranked as top quality because the player isa ble to win regardless.

Sorry for the silly analogies, but I hope they bring out my points a little. Net Dmg per minute relative to the average ranking or the individuals own team vs the average of the opponents team should yield very accurate results of the real skill of the individual. But if you say you've already tried this and it didn't work, then maybe there's something that needs tweaking?

Slightly off topic: I would love a feature where we could filter the server list according to average elo of the servers, any chance this might see the light of day in the future?

Keep up the good work :)

The problem is you seem to think damage is really important (you aren't the only one). It isn't - winning is solely important. Ego is clouding your judgement.

"The system as it stands is completely broken. It should be based on dps while alive/total damage/how you placed compared to the rest of your team."

This is the problem with many CA players. They really want to be playing a single player game - this guy wants to be ranked based on BEATING his own team mates (more damage, living longer than them, being higher in the team scoreboard than them)..

For a 1v1 gamemode like duel winning is solely important and allows the elo formula to avoid all the complexity involved in team game rankings. I think - and I don't mean this as a flame or anything - that you are making a fundamental conceptual mistake in applying the win loss approach to the team game model.

In team games, although you are temporarily aligned with your team mates for a single match, on the rankings table you are direct competitors with them, so your discrete performance as an individual needs to be taken into account relative to both your teammates and your opponents in order to generate your ranking. By comparing performance both against team mates and opponents you also almost double the data informing the scoring formula which should correspondingly enhance the formula's accuracy over time.

Win/Loss relates to team performance in a team mode, but the rankings relate to individuals. Dmg/Frags can be directly correlated to an individual's performance and hence be a more accurate measure for ranking an individual player on QLRanks.

Let's flip this on it's head for a second to take a different perspective:

1) The accuracy of a ranking formula can to some extent be measured by the degree to which it is prone to erroneously score an individual's performance.
2) The win/loss formula can both result in poor players getting +ve elo points if the team wins, and good players getting -ve elo points if the team loses.
3) The Dmg approach cannot mistakenly take poor dmg output as an indicator of good performance, nor can it take high dmg output as the indicator of poor performance.

Therefore the Dmg approach is a more accurate measure of an individual's performance. And since the QLRankings relate to individuals and not teams, then this approach should be favoured over the win/loss formula.

Last example: do you think Lionel Messi's transfer fee is determined by Argentina's loss at the world cup final, or by his individual performance in every match that he plays?

In my opinion you're wrong Sirax.

You are ranking INDIVIDUAL players with TEAM results. If you put (for example) ic-TOX on a team of lower skilled players vs a great team of players, he will INDIVIDUALLY be ranked based on his TEAMS result. He would accrue a _terrible_ ELO even though he's one of the best players in the world, and may have the BEST damage/kills on either team.

Also what about tier 3 players having the same elo as tier 4 players etc?

ps - total respect for the work you've done btw, I'm just trying to get my point across.

I can't keep having this same conversation.

Your tox analogy is nonsense, I can't be bothered to explain why.

Trying to shoe horn 'tier 3' 'tier 4' is nonsense, you're trying to compare two rating systems. Tier 3 and Tier 4 players might have comparable ratings with QLRanks because, shock horror, we don't recognise the difference between tiers because, shock horror, the tier system is bunk.

I understand you're passionate and think that someone doing loads of damage is important. It feels that way as you are playing. It is important on a small scale too. STATISTICALLY it just isn't useful for calculating skill on a grand scale.

Why do you think our top 10 is arranged the way it is? It's because those guys make their teams win. They make their teams win because they are the best at playing CA in 4v4+ mode. That doesn't mean these guys do the most damage, it means they are good enough players that they compensate where necessary. They identify the correct things to do that make teams win.

Look, you sound like you want someone to be compensated ratings wise for doing what you perceive to be good play when they are losing/when the teams aren't fair etc. The reality is the law of averages solves this problem on a long time frame. If these guys are doing tons of good stuff, it will be working more often than not. If they are doing good stuff but still losing and not gaining rating - maybe their good stuff isn't that fucking good! This is why the top 10 of CA looked awful in 2012 and now it looks good.

Relax. Play Quake. Stop getting upset when you lose rating.

When people who play CA on all european servers, several hours a day, for many years tell you that the current algorithm is rubbish then it must be the case.

Most of us do not care about your ranking, however it is solely based on winning-losing and it makes people camp, rage, quit, cry, troll and in general, it ruins CA since it was introduced. It makes people register fake accounts which discourages tier 1,2 and 3 players. Once they reach tier 4 it ruins shuffle for a while before they reach the adequate skillrating.

You argument about the top is absolutely invalid. Teams are shuffled according to skillrating not your elo. Many people reach 400 skillrating and as far as the game is concerned, are just as good as your top1. It is often the case that despite many shuffles teams will never be even.

You refuse to think logically. It would make sense if it was a clan instead of individual players that would receive elo points. That is how all ladders work. This is why if you want to rank and award players individually, NOT as a team, then you must take into account their individual achievements, such us damage. Therefore only duel and FFA can be appropriately ranked following your current way of thinking.

It is your site, your ranking, your rules so you can make it whatever you want it to be. But please stop being delusional.

Don't get upset (because that is a nice way to end a post).

Try to look at it from my point of view.

I'm an engineer who bases his decisions on logic and mathematical evidence. This algo has a mathematical paper that i've read (you haven't).

I have no specific attachment to it. I didn't design it, test it or invest in it. I did make my own first (based on dmg, position, etc) and it was inferior.

If you're going to complain about it, you need to bring me figures and details. Tell me WHY it's bad with REASONS and EVIDENCE.

'Anonymous from Ireland' saying 'It's rubbish cos I say so' is not evidence.

'Exp should be above Teuz4' is not evidence.

'DMG is rly important bro' is not evidence.

I look down the list of the top 100. It looks accurate. I look down the list of the top 100 tdm. It looks accurate. I look at Exp and Teuz4 - it looks accurate.

You're complaining but I cannot work out what exactly is the problem. The problem I think is you just don't like being ranked, which is fair enough. That's not a very good reason to call me delusional though.

So you needed a paper to develop 40*(RATING - (ACCURACY * K)).

More comprehensive reading required. My second paragraph gave you reasons. What evidence are you talking about, do you seriously expect anyone to go so far to prove you wrong? If you cannot understand it yet then you never will. Sorry Doctor of Quake but I'm not writing the PhD thesis for you ;).

Are you dense? That's an equation used to stop massive jumps in rating due to the confidence / rating variables of trueskill.

https://research.microsoft.com/pubs/74419/tr-2006-80.pdf

Is the research paper I was referring to.

As for the rest of what you wrote - you're just proving my point. Whine without substance.

"Delusional" fits you perfectly and you are as stubborn as a mule.... People try to tell you that the system that only accounts for winning-losing is not ideal, especially in quake where there is little variables (no recoil, no headshots). Microsoft uses the same system to then shuffle the games. Yours is separated from the in-game environment.

If you are an engineer then why do you choose to ignore what happens in practice. People rage over elo. If you would try to abort a game before the end there would be suicides.

Most people do not care if there is a ranking for public games. But they start to when it impacts everyday playing. EOT.

People complaining a lot about something doesn't make it right. Elo's point is to measure performance, nothing else. It's not here to tell you how talented you are or to make you happy.

What we have in this thread is classic clash of people's expectations with reality.

I don't agree with that.

The ranking system might be the most appropiate in terms of ranking players, I can give sirax that by all means.

The problem is that the ranking system has such focus on winning which in turn creates a pattern among player behaviour in terms of choosing which teams you join and at which point. I know that a lot of the players in the top 100 deliberately don't join teams they know are inferior or games where one of the teams is losing heavily. The community isn't big and most players have a pretty good knowledge of how good everyone else is.

In maybe 70-80% of the games I can tell straight away which team will win (unless someone quits or goes afk). I'm quite confident others can too. I won't call out any names but pretty much everyone in the top 100 picks which games to join to a varied extent. I know it cause i've seen it. I'm sure others has too. The odd guy being k1llsen perhaps since he doesn't care about the rating and he's also way above everyone in terms of skill by quite a margin. The argument that they're so good that they will win anyway no matter which team is bullshit.

The ranking system works fine until the point where everyone joins a game, shuffles, smiles and just plays. But this isn't the case.

The percieved reality of the ranking system does not clash with the actual reality. The behaviour of picking when and who to play is most likely a combination of players naturally not wanting to lose but also being aware of when to pick your fights in order to maximize your chances of getting ELO. Most don't know anything about how the ranking system works but have a strong notion of when they usually win rating and that's all it takes to end up with a ranking system that is pretty good but far from spot on.

How to solve it? No idea but taking the ranking with a pinch of salt seems like the current option for now.

Ok - I understand now what the formula aims to achieve. It successfully identifies and ranks those players that are so extremely good or extremely bad, that their presence on a team will most likely make it win or lose. In that it appears to be successful. In providing an accurate ranking system for the thousands of players that don't fit into that category, however, it seems a little skewed.

Anyway, I won't bother you anymore, I just thought you might appreciate some player insight, but I understand if that is too irritating for you.

Keep up the good work and thanks.

The point is that someone always affects the teams performance, regardless of their skill level. The amount they affect the team performance is measurable over the long term. Ranking CA requires large volumes of games to be accurate, unless the effect is especially stark - they are so good that they almost always win regardless of team balance.

Any inaccuracy exists because the volumes aren't there - that's why we have a confidence variable which aims to smooth that, but it does make people complain. A confidence shift can make your 'rating' go down, even though your actually strength variable goes up. The side effect is that as someone improves it will take a while for their rating to reflect that accurately - but it does happen.

Believe me, if a real mathematician (i.e. czm or equivalent) came along and had some intelligent criticism I would be the first to bow to that knowledge and invite their corrections. The problem is that whilst there are many many people out there capable of that the scale of work required, it would be approaching thesis level in complexity and scope. That time investment is better suited to other things for most people.

I realise that I can come across as arrogant regarding this stuff, but the reality is that you are not operating on the intellectual/mathematical level to really contribute to improvements to the algo. The fact is - I am not either. I tried to write my own and it was just rubbish.

There are active suggestions to be made regarding the handling of game numbers, quitters etc. These were metrics that I went through with Lam about 12 months ago to try and improve the situation. They could be further improved.

The algo itself however is pretty much off limits, unless you're waving your double major in maths/statistics coupled with CS. Then you're free to produce an improvement and I would be excited to see it.

I appreciate that the maths must be challenging, but success of the maths must yield successful real world results. Myself and others are just providing the empirical evidence to help evaluate the success of the maths. We don't need maths degrees for this. But if what you're saying is that although it is not perfect at the moment, it is still broadly accurate over time. And that the work to improve the granularity of the ranking algorithm is too challenging to be worth the effort for some slight improvements, then that is perfectly understandable.

Yep.

I'd still really like to see examples where the algorithm has clearly failed however. I think it's accuracy is pretty damn impressive. I'm more impressed by it now than I was 3 years ago when I originally implemented it.

I thought a win based algorithm was way too simplistic. I was wrong.

Any match where one player has played amazingly well, done incredible amounts of frags and dmg, but narrowly lost due to the fact that the rest of their team were a bunch of nubs is an example of where the algorithm isn't working. Because despite the amazing performance it does not add to his ranking score.

I don't want to go over this again - it's just you are asking for examples and there are thousands.

He's not asking for a single match. Or even for thousands of matches for different players. The law of big numbers takes care of that happily. He's asking for examples of players being ranked higher than they should along with an explanation of why they shouldn't be ranked this high.

Well I don't really pay too much attention to others' elo progress but I do notice that there are many 2000+ elo CA players I own regularly when I am 200 points below them, whereas there are 2000+ elo players who I just can't touch they are so good. I'm guessing the former group have high elo due to their quitting to avoid losses.

That aside the point I'm getting at is that there is a big gap in skill between the players that are good enough to constantly end up on weaker teams (I know the shuffle is done on QL ranking system which is different) and those that are good enough to carry the entire team consistently enough for the QLRanks algorithm to register their superior skill. It's like a microphone that only records when you shout at it, but doesn't register a thing when you speak loudly to it and treats your voice like background noise :)

elo doesn't work on a game-by-game level, but over a long period of time it evens out between the times your team was good and the times your team was shite.

Problem is there is no empirical evidence in this thread, just individual perception.

On servers with minobot the matches are more often than not very even. Exceptions occur precisely when a player doesn't have many games under their belt or is rusty/having a bad day (or a good day), which the algorithm takes into account, as elo you lose (or gain) because of a bad (good) teammate will then be re-gained (re-lost) in further games.

The two major criticisms I have of the qlranks elo are:
1) It doesn't separate PQL and VQL elo, which is quite problematic for people who play both.
2) It doesn't do anything to adjust for late joiners. I don't know if some kind of factor of time played/total match time could be added to adjust for elo change in games where someone joins the losing team when the score is 9-1 or something (my statistics course is 2 years from now). I'd say the reason lower tiers have such inconsistent elos is precisely because of how many quitters there are in those tiers and how those mess up the calculations. Someone who should win a best of 10 easily won't necessarily win a 9 round deficit.

Also Sirax, how many times do you see in CA a player of really high skill on a bad team, this player does amazingly well / gets the highest score on the server / makes his team win rounds they had no chance of winning etc etc, and the system takes ELO away from him?

This is the reason people quit bad teams / don't join / slum etc. How can this be a good system? If you took the advice of all the other people recommending changes, these issues would not exist (more individual based results)

No, I said to also factor in dps while alive.
The system gives me ~1600 ca elo when I can beat any of the other ca players in 1v1 CA, and I consistently place first with significantly more damage.

e.g: http://www.quakelive.com/#!profile/statistics...1fd90/ca/1
or more specifically this:(jja, wraith, end) (or any other game on my profile really) http://www.quakelive.com/#!profile/statistics...1fd90/ca/1
They have significantly higher ca elo than me yet they have accused me of hacking.
I don't think that's how an elo system is supposed to work.

Not that accuracies mean too much, but I find it hard to believe a player with 47% lg and 60% rg as 1600 elo. I simply don't care to teamstack, I just try to make sure I am doing as much damage as possible. The current system penalizes me for this.

It is not my fault if you are too blind to realize this, which is why most people ignore CA elo altogether.

On average, their team is made stronger with their presence then with yours, where "stronger" is a mathematical equivalent to "more likely to win". Therefore they are ranked higher. Seems reasonable to me?

"but I find it hard to believe a player with 47% lg and 60% rg as 1600 elo"

I find it perfectly believable for a 1600 Elo ranked player to hit those accuracies in at least one game.

By the way, if there team was had an higher Elo than yours (as you seem to believe) then you must have gained Elo for the win (and similarly they would lose Elo). The system works!

That is my average accuracy... I have several games where I've had 55%+ lg and a few with 60%+

e.g http://www.quakelive.com/#!profile/matches/Ob...1fd90/ca/1
http://www.quakelive.com/#!profile/matches/Ob...1fd90/ca/1

etc
Since I lost the 2nd game, I lost elo. Does that make sense to you?

You've played 56 clan arena games according to your profile and your CA Elo has an upward trend. I see nothing unusual here...

But then look at this person's profile.
http://www.qlranks.com/ca/player/rawkits
He never scored much higher in any of the games, he just decided to pick the winning team when he joined.

His win rate is 78% versus your 50%. He was assumed to be 1200 Elo but he kept winning so the system quickly adjusted him to a much higher score. You only won 50% of your game so you didn't rise as quickly. I still don't see a problem.

In response to your edit: now we really need sirax to step in, but my understanding is that the system only counts reasonably balanced games, for some definition of he word "balanced". How do you know he just joined the winning team? You mean he waited until the match had started and joined the team that was most likely to win?

You can easily join the winning team. Just stay in spectator mode and then join the team with the better players once someone leaves.
If you don't like your team you can just spectate, a lot of people do this.
It should be based on how well you are doing, not your win rate. This is easy to abuse.

Yes game balance is a problem. But that isn't what your initial criticism was.

You were saying that the rating should be based on damage which is just not wise. The point is that damage isn't a metric that on large scales indicates your worth to a team. Winning is that metric, whether you like it or not.

The win joining type stuff is an issue- but I'm not convinced it's causing a big issue on the ratings board. It's not like someone has win-joined their way to the top 200.

This is how these discussions always go. Someone wades in going on about damage, I contest their points, they then mid-argument switch to some other nebulous point (win joiners, quitters, whatever). Eventually settling on calling me stubborn.

Come with one point. Prove that point is in issue.

Then we can make some progress.

I don't see how damage wouldn't be the main metric for a team. If you assembled a clan arena team for example in a 3v3, you would want the best aim players who will be doing the most damage.

I will leave it as I don't agree with your metric. In the top CA rankings for the United States none of those players are very good except for a few like : nemesis, kgb and such.

However, the vast majority of the rest of them all think I cheat. I don't considering them very good players, but qlranks says otherwsie. I know a few players who have beat them 10-0 in 1v1 clan arena, but qlranks doesn't count 1v1 or 2v2 towards your rank for whatever reason.

For whatever reason? It's simply because it's trying to evaluate the skill of people playing a team mode. With teamwork. It's not trying to evaluate who aims the best. The best aimer can also be a bad CA player.

joining the winning team isn't as easy as you are saying.
how often does it occur that someone leaves the winning team and there is a free slot? its pretty rare.
plus who is stupid enough to waste his time like this rather than actually playing the game for long? .. no one!

there are so many people like you who are good individual players who do tons of damage and have lots of kills.
but on the other hand these type of players are always late with the whole team (because to increase your chance of winning you usually meet up at the start), let mates die so they don't and don't stay at one spot to cover mates backs .. in short they wont ever sacrifice themselfes even if it would increase the chance of the team winning.
and in the end they tell you "nab team, scew you all. plz do dmg."
they never suggest what to do as a team after the first round is lost such as "let's meet at XY" or "focus ObiousAlias" for example.

if you want an ELO system that is based on how well one is doing, why arent you playing duel?

Nobody plays duel on the west coast of the United States. I can sometimes get a few games with 90 ping if I play on the East Coast, but Quake Live's netcode is not well suited to those pings.
Duel is only really popular in Europe.
The aiming between CA and Duel is very different eitherway, but I don't mind.

No. You're just assuming that me (and others) believe that it is the most important thing. We don't assume that you believe that the team winning is the only thing that matters. What we do question is why the ranking system is based purely (perhaps it's just heavily) on that and not a combination of winning/losing and the other so-called arbitrary figures.

So what have you already tried? :)

I understand that a ranking system can get very complex and there's many different ways to look at. I know that it isn't an easy thing to do and after all it's just Clan Arena. I just get the feeling from this and past discussions that you're not really willing to discuss it with the community or change anything in what looks to be a ranking system for a specific game mode that many are questioning.

Edit: Sorry, I thought it was based only on winning/losing but i realize that isn't the case. Sorry for that. Well, at least I understand that the ranking system boils down to what you think is the most important thing. Perhaps you are right that winning should be the most important thing. Perhaps there should be slightly more focus on other aspects. I'm glad that there's a discussion about :o

You used to lose a lot of rating for quitting, but was reduced a lot. Not sure how much it is now.
But the system works quite fine, like sirax says: "play more quake", and it will be accurate.

People over 2k elo dont get there by accident, if they have played a lot of games..

Thanks. Well, what can I say. I believe that the current ranking system maybe puts too much focus on winning which means that if you're only looking to get ELO you shouldn't be taking any risks by joining any team that is losing the game. It means that "play more quake" is partly right but who you play against and in which situations you decide to play has a pretty big impact on your ELO.

On the other hand if you put too much focus on invidiual performance (damage/frags) you end up with everyone forgetting the team aspect of the game since it doesn't really matter if you lose or not as long as you do decent.

Perhaps ELO should be handed out as it is right now but also take into consideration the total amount of ELO each team has and the amount of rounds won by each team along with winning/losing the game? :)

It does take into consideration the "total amount of ELO each team has"... well not exactly that but a similar team skill estimation value. TrueSkill is similar to Elo in that regard.

How come it uses a different skill value other than ELO?

So I googled that and ended up with the skill rating and confidence formula. Isn't that only used to calculate the players ELO and not to influence the ELO won/lost based on the teams total ELO compared to the the team you're playing against?

Both teams have their rating calculated based on the rating of the constituents of the team. The outcome is based on the disparity between the team ratings.

i.e.
'Perhaps ELO should be handed out as it is right now but also take into consideration the total amount of ELO each team has'

Rating has always considered the team rating.

Thanks!

Why is it possible to get minus elo for a win? I mean, it seems kind of silly to get minus rating for winning... (no matter if it's a 10-9 or a 10-0 stomping).

It punishes a player for random public matches that have bad shuffles etc. When I decide to play I can't be arsed to first check servers to see if the 'average elo is high enough' to make sure I don't get minus elo even in the case of a win.

Plus, after map change people leave, new ones join, etc. Should I then every time disconnect to find a new server again that has high average elo (which at many times of the day don't even exist)?

What's the reasoning for not making it so that a win will never give you less than +/- 0?

Why punish someone for winning if you think that winning is the best/most important measure of a team game? People can't stat whore by playing vs. only noobs if they end up getting +0 every time. But why punish a top player (like k1llsen) who joins a public server with his 2700 elo for stomping the other team 10-0. It's not like he has influence over what the average elo of the other team is on a public server.

http://www.esreality.com/post/2372668/quakeli...pid2619566

Any inaccuracy exists because the volumes aren't there - that's why we have a confidence variable which aims to smooth that, but it does make people complain. A confidence shift can make your 'rating' go down, even though your actually strength variable goes up. The side effect is that as someone improves it will take a while for their rating to reflect that accurately - but it does happen.

No it isn't possible, but it seems that way.

Explained here:

http://esreality.com/post/2619922/#pid2373392

How about making it more accurate by having it identify and remove deleted accounts!? :

eg

http://www.qlranks.com/ffa/player/NEPHlE_CC

just sits in UK top 10 for 4-5 months despite being banned, then deleted.

This highlights another problem, there isn't some kind of decay system too, as there are quite a few Elo hunters that rank up a high score, then just abandon that account. It doesn't ever drop off no matter how others improve.

And how about at the very least suspending banned accounts too (store their current elo, temp swap it to 0, hide them from the site), as some of them are just cheating so it's not a temporary situation.

@vanne
Hahaha Ikea boy, I saw u ragequit in the middle of game multiple times so u beter stfu

quality post award goes to you

CA whiner award goes to you

Says the guy who literally cried in every game I've ever seen you in. Oh - there's also nothing more enjoyable than seeing you ragequit, silverthorn. :)

I have a faint memory of this silverthorn guy being one of those odd assholes that you come across every now and then that takes whining and raging to a completely different level. Most whiners are slightly annoying but the odd ones just puts a smile on my face :D

Most of them keep it in-game but this dude delivers considering he only has a handful of posts here on ESR :/

Yep, he's definitely one of those. :)

they (not qlranks) should rather focus on fixing the team shuffle.
i and many others care shit about losing a game (and therefore "elo"), but i do care much about having a good and balanced game (as far as this is possbile with the given players).
its just too random.

ontopic: the CA elo rankings feel relatively accurate imo.

Yes I feel its very well ranked, especially if people play a lot.

ELO/IDIOT

QLRanks. Taking LOLCA to new levels of LOLERS.

I noticed with many players that the ELO on top difference from the ELO at the bottom in the World Rank box.

But that must be me seeing things...

One of the many examples - http://www.qlranks.com/ca/player/1NJUSTICE

Yeah in the world rank box you always have kind of the average of all players of the box as elo

people still take qlranks ca rating seriously?
elol

This elo is just stupid, works only in duel.

lolca

lolCA elo

why skillrating 20 duell noobs keep calling lolca (((

lolskillrating

I just find it funny that the OP's QLranks link shows he hasn't played in 400+ days, so around the time he made this post :p

what is an alt account?

dingdingdingdingding

tdm is even more retarted.
i lost 2 games and got -500 elo lol

your elo is nothink on the CA

Started reading this thread and started writing my own comment.
Until I came across sirax's explanations and realized: 'ok, I better stay out of this'. ;)

For those who feel like they are being wronged by the ranking: maybe you can take comfort in the fact that the algorithm is the same for everybody else. So even if it seems unfair to you, it's going to be unfair in the same way to everybody else. So it should all cancel out in the long run.