If you add "serious" in the thread name it won't make it serious

The only place it works is reddit, and there only because it signals a harsh modding standard on the subreddits it's accepted.

everyone watch out , we got interwebs connoisseur here

You're just envious.

X D
D

"If you do not care about FFA ELO, this discussion is probably not for you."

Phew.
Ok hf.

The same problem occurs within in the CA elo as well. Whether it can be countered, is another problem.

We don't use ELO on qlranks for team games, it's just poorly labeled on the website.

And you gained my eternal hatred for your use of "Elo" about things which aren't Elo!

Seriously, that triggers my OCD every time.

I know, I should probably change it - it really is horrible. I guess my point was that everything this guy wrote is irrelevant to how qlranks calculates rating changes.

People really get upset when the frontend says they lost points for winning. It's really because of a shifting confidence factor that isn't exposed to the frontend user..

I guess my point was that everything this guy wrote is irrelevant to how qlranks calculates rating changes.

I'm sorry, but am I the "guy" you are referring to?

clearly

Alright let’s proceed based on that assumption.

So Sirax, your response to thoughtful & constructive criticism on how QLRANKS.COM implements FFA ELO is: “everything this guy wrote is irrelevant” ?

An interesting and perhaps telling response, but I beg to differ.

After some modelling, I have concluded that QLRANKS.COM uses a valid algorithm for calculating DUEL ELO. However when analysing changes displayed for FFA ELO, I have concluded that QLRANKS.COM does not use a valid algorithm to calculate FFA ELO.

Here are 5 observations which support that claim.
I, and perhaps others, would be interested to hear an explanation of why our observations of QLRANKS.COM’s DUEL ELO and FFA ELO per-game changes differ so noticeably on these 5 points.

I then went one step further.
Instead of simply complaining about how QLRANKS.COM calculates FFA ELO, I researched the problem and formulated an alternative algorithm. An algorithm which may not be perfect, but which addresses all 5 problems highlighted above.

I created a working prototype of the algorithm for demonstration purposes, and finally posted an example of the logic codified. All of this is freely available for you to use to improve QLRANKS.COM , no charge.

So Sirax, please let everyone know if you stand behind “everything this guy wrote is irrelevant”.

Everything you wrote is irrelevant because you don't seem to realise that the terms 'elo' and 'rating' aren't synonyms.

Fair Enough, so your position is:

QLRANKS.COM uses the Elo rating system for rating DUEL matches
QLRANKS.COM uses a proprietary system for rating FFA matches
Correct?

But your webpage is not 100% consistent when applying these labels
FFA rating (Frontpage) labelled Elo
FFA rating(Top in Country) labelled ELO
FFA rating(FFA page) labelled Rating
FFA rating(FFA player page) labelled FFA Elo

So you can see how easy it is to assume that 'elo' and 'rating' are synonyms within the context of QLRANKS.COM

However, I am happy to agree that all your “proprietary FFA rating system” shares with the Elo rating system is a similar range for scoring players (approx. 800 - 2600).
But changing the label does not remove the root cause of my criticism.

It is my position that your current FFA rating system performs poorly, is not very popular, and should be either scraped or replaced.

Assuming you feel differently:
Please tell us what you perceive as the positive attributes of your current FFA rating system.

See comment 16...

Sup now faggot?!

See comment 16...
Sup now faggot?!

While I wouldn't personally word it that way, I agree with the sentiment.

Sirax: Thus far you have contributed one single argument to this conversation…..which in itself was just an excuse, a justification for you not participating in this conversation.

Now Comment 16 completely demolishes this one semantic distinction you tried to shield yourself with.

I am aware that this is a discussion you are trying to avoid, to you I am a nuisance you’d rather do without. But I have a valid point, and I have no intention of dropping the subject.

So if you could do us the honour of typing more than one sentence at a time, we’d be keen to hear if you wish to defend your current FFA rating system, or if you are open to improving it.

Maybe go tell the Microsoft research team who developed TrueSkill that their algorithm is "invalid", since I think that is what is being used.
http://research.microsoft.com/pubs/67956/NIPS2006_0688.pdf

edit: skimming through the paper, this section might be of interest to you (page 6)

We compared the TrueSkill algorithm to Elo with a Gaussian performance distribution ... When we had to process a team game or a game with more than two teams we used the so-called duelling heuristic: For each player, compute the delta's in comparison to all other players based on the team outcome of the player and every other player and perform an update with the average of the delta's. The approximate message passing algorithm described in the last section is extremely ecient; in all our experiments the runtime of the ranking algorithm was within twice the runtime of the simple Elo update.

and later:

... we see that TrueSkill is signicantly better than Elo for both the Free for All and Head to Head game mode but fails in Small Teams.

... As can be seen, TrueSkill automatically chooses the correct learning rate wheras Elo only slowly converges to the target skill

So it looks like TrueSkill is better than your proposed system and the problems you've pointed out aren't actual failings but artifacts of the fact that the system displays a conservative estimate of your actual skill (given by mu - 3*sigma) combined with this:

If the skills are expected to vary over time, a Gaussian dynamics factor N (s_{i;t+1}; s_{i;t}; gamma^2) can be introduced which leads to an additive variance component of gamma^2 in the subsequent prior.

Finally, I think you need to remember that the absolute number doesn't matter. It's a ranking system, only your score relative to other players matters and statistically this system has been shown to outperform your "duelling heuristic". Why should we sacrifice accuracy just because some people get pissed when their absolute numbers occasionally go down after a win?

This whole Elo strife the forums has been infected with is just stupid.

7/10 obi, for contributing meaningfully and intelligently to this conversation.
Give me a day to review everything that you have written and linked to, before I am in a position to give a proper response.

-- Edit: lowered Obi's score --

Obi, I’ll admit the article you linked to is a tough read. If you understand it, you are smarter than I am.

But the Wikipedia article on TrueSkill is rather manageable.
So I’m going to borrow a page from Sirax’s playbook: I’m going to sidestep the argument you are proposing (TrueSkill is better than Multiplayer ELO), and instead challenge the validity of the premise (That QLRANKS.COM is using TrueSkill)

If the Wikipedia article on TrueSkill is accurate, then I can use it to construct several arguments against the probability that QLRANKS.COM is using TrueSkill.

1.) Wikipedia states: “TrueSkill is patented,[3] and the name is trademarked, so therefore it is limited to Microsoft projects and commercial projects that obtain a license to use the algorithm.”

I very much doubt that QLRANKS has the funds to acquire such a license, or that Microsoft would provide it free of charge given that they are competing for online gamers.

2.) Wikipedia states: “Published formulas for Trueskill are not complete, and some needed functions are simply shown as graphs. Some of the math is spelled out in the paper.”

If this statement is accurate, then we cannot actually gain a complete understanding of exactly how to implement TrueSkill, even if we understood everything in the article.

So how did Sirax and szr fill in the blanks to be able to encode the logic?
Maybe they are smarter than I am, and spent more time on the subject then I have.
But I personally would bet money against this being the case.

3.) Observed behaviour vs Expected behaviour.
We know that on QLRANKS.COM you can lose rating (and ranking) by winning a FFA match.

Nothing I was able to extract from the pdf article can begin to explain this observation.

Wikipedia states: “On Xbox Live, players start with mu=25 and standard-deviation=25/3; mu always increases after a win and always decreases after a loss”.

This sentence would indicate that On Xbox Live, when rated by TrueSkill, a player must lose a game to decrease his rating.

Finally, I think you need to remember that the absolute number doesn't matter. It's a ranking system, only your score relative to other players matters..

I agree.
And I am in no way complaining that my FFA rating or ranking are too low. I am complaining that the way QLRANKS.COM adjusts ratings after a game does not appear follow even basic logic.

Have a look at this page
I have compiled a few examples where Winning a FFA game resulted not only in a loss of rating, but also a loss of ranking. In each example I traced the rating change for all players. If you can explain how the ratings adjustments shown in these examples “increase accuracy”, I would be very interested.

Sighting these examples, I still believe that what ever algorithm QLRANKS.COM is using for FFA ratings is deeply flawed and should be scrapped or replaced.

--Edited for clarity --

So I’m going to borrow a page from Sirax’s playbook: I’m going to sidestep the argument you are proposing (TrueSkill is better than Multiplayer ELO), and instead challenge the validity of the premise (That QLRANKS.COM is using TrueSkill)

I just asked him and it does use TrueSkill.

1.) Wikipedia states: “TrueSkill is patented,[3] and the name is trademarked, so therefore it is limited to Microsoft projects and commercial projects that obtain a license to use the algorithm.”
...
2.) Wikipedia states: “Published formulas for Trueskill are not complete, and some needed functions are simply shown as graphs. Some of the math is spelled out in the paper.”
...
So how did Sirax and szr fill in the blanks to be able to encode the logic?
Maybe they are smarter than I am, and spent more time on the subject then I have.
But I personally would bet money against this being the case.

The Wiki is out of date, they made the full algorithm public (in the paper I linked). Also, you find it hard to believe that they are smarter than you? Lol ok, why do you feel the need to bring this up?

3.) Observed behaviour vs Expected behaviour.
We know that on QLRANKS.COM you can lose rating (and ranking) by winning a FFA match.

Nothing I was able to extract from the pdf article can begin to explain this observation.

Wikipedia states: “On Xbox Live, players start with mu=25 and standard-deviation=25/3; mu always increases after a win and always decreases after a loss”.

This sentence would indicate that On Xbox Live, when rated by TrueSkill, a player must lose a game to decrease his rating.

This is expected (perhaps undesireable) behaviour and the reason can be found in the paper (go figure):

The TrueSkill skill of a player i is currently displayed as a conservative skill estimate given by the 1% lower quantile mu_i - 3*sigma_i. This choice ensures that the top of the leaderboards (a listing of all players according to mu - 3*sigma) are only populated by players that are highly skilled with high certainty, having worked up their way from 0 = mu_0 - 3*sigma_0.

If the skills are expected to vary over time, a Gaussian dynamics factor N (s_{i;t+1}; s_{i;t}; gamma^2) can be introduced which leads to an additive variance component of gamma^2 in the subsequent prior.

Honestly it's bad form to argue from Wikipedia when you have the full paper in front of you.

I have compiled a few examples where Winning a FFA game resulted not only in a loss of rating, but also a loss of ranking. In each example I traced the rating change for all players. If you can explain how the ratings adjustments shown in these examples “increase accuracy”, I would be very interested.

Using individual results is meaningless for determining the performance of a large-scale, statistical algorithm. The paper clearly shows the advantages of TrueSkill compared to Elo.

I'm going to give your comment 2/5 potats.

Well how do you feel about losing points for winning? What's wrong with +0, at least for team games where you can't control the teams? That way there's still nothing to gain for roflstomping noobs, but you also can't lose anything for actually playing.

>Today you can win a FFA match and still loose ELO points!

yes if you win versus low tier players

When i come first i lose elo even with 8 people playing and the gap between me and second is like 10.....gg wp

What makes FFA ELO unreliable are the powerups. They wildly alter the match. Some players can't win without them, some players don't even need them, but they still change the outcome in one way or another.

is that so

Powerups don't make ffa random. Ffa is random by nature of it being ffa. What's the.point of me constantly timing red if I have a 50% chance of getting back to it before it respawns? Not to mention its all kill stealing.

lol who cares about FFA ELO??

0/10 didn't read OP

the FFA "ranking" system is garbage. nobody should ever lose points for winning a game. I don't care about your "unexposed shifting confidence factor," whatever that means (how about explaining it instead?).

fix your shit.

Pretty sure he provided a link to the research paper with the full implementation. Or you can just look up TrueSkill on Wikipedia.

Just stop sucking dix in tier3 servers \o/

^^^ mad, cross eyed mongoloid

^^^ mad, cross eyed mongoloid

Nice thread