see PredaTh0r's comment here http://www.esreality.com/post/2789202/re-curr...pid2789202 for about all that's been said on a forum, and where to go to talk about it.

TL;DR: DO NOT EVER TRY TO COMPARE RATINGS FROM DIFFERENT SYSTEMS.

qlstats.net is an exerimental prototype.
Nobody knew what anomalies are slumbering in the QL match results, and the only way to find out was by collecting data.

At this time data is mostly unfiltered and i'm working crazy hours getting the code in place to recalculate everything. This is a lot of work and doesn't happen over night.

The current system is based on traditional Elo, using Score/Time as the metric to decide the winner of each player pair and adjust the ratings. This works quite well for all game types except CTF. With "well" I mean despite all its flaws and the garbage in the database, games shuffled using these imperfect ratings are more likely to produce balanced teams than a random number generator.

I'm in the middle of switching over to Glicko-2, which is more similar to what qlranks had. But again, don't try to compare apples and pears. Unlike QLranks I will continue to use individual player performances and not team win/loss for the rating adjustments.

And I'm considering to hide the Elo changes from the match results. People get obsessed with numbers which they don't know how to interpret. They are needed for matchmaking, but require some understanding of statistics and the system (e.g. "why do I lose points even when I win").
An "unlock Elo column"-questionnaire would be nice, but I don't have the time for it.

Man, you're doing great work. Props to you!

great work

imo hide elo from team games and display it in duel (and maybe even ffa), since in those it helps u find better games, measure ur performance and progress.

imo elo is perfectly suited for duel and it is something used in other 1vs1 games also, like chess.

but elo for team modes, isn't a good idea. At least, it should never be displayed. It should just be an internal number helping for more balanced games and that's it. No offense, but u can't be playing CA, which is a mode created for fun and want to brag about elo. You either play for fun, or u play competitively in competitive modes. Having elo in team modes, which to start with is really hard to measure since u r not playing alone, results in obscure behaviors and pathetic abuse. We all have seen that one poor guy, who was the worlds No1 in some if not all of ca, ctf and tdm modes.

tl;dr elo is good for 1vs1 games, but not for team games and that's why u don't see it in any team games IRL. For sure, it can and it does help to have more balanced games, but it should never be displayed.

Rating is a complicated topic, so it is natural to get a few things wrong. I also had lots of wrong assumptions before I started looking into this. My math will never be good enough to fully understand it, but I think I got the concept now.

Elo and Glicko can very well be used for team games.
As long as you can calculate a metric of each player's individual performance, you can order players and apply pairwise win/loss comparisons to adjust the rating. The magnitude of the performance metric has no influence on the magnitude of the rating change, it's only used to sort the players by performance.
The amount of the rating change depends on your own and your opponents current rating and confidence in those ratings (mu and sigma in a gaussian distribution).

[EDIT]
I just realized that this pair-wise (although often used) has its own pitfalls and anomalies. That's where the math ends and dirty workarounds begin.
[/EDIT]

With time and the number of matches, the initially unsorted playerbase gets more and more sorted and each pairwise comparison improves the quality of each players rating.

What these systems do not support is rating the team as a whole and then distribute the team-win/loss back to each player. This is NOT what qlstats is doing. This is what Trueskill can do. But Trueskill is patented and all the javascript implementations I found produce incorrect results. The python implementation seems to be correct and the overall ranking of players (top 20) has more or less the same players on it as Glicko2.

What Glicko and Truskill have in common is to use 2 numbers for your rating. The estimated rating (r or mu) and the uncertainty (rd or sigma). Over time the uncertainty usually decreases when your performance is constant.

Here are some good reads:
Elo vs. Glicko: http://www.bjcox.com/?p=120
Glicko: http://www.glicko.net/glicko/glicko.pdf
Glicko2: http://www.glicko.net/glicko/glicko2.pdf
Truskill: http://research.microsoft.com/en-us/projects/trueskill/

I don't have any math knowledge but maybe this can help http://smashboards.com/threads/skillkeeper-tr...er.376590/

Thanks, didn't know about that.
But it's using Trueskill, which is patented by Microsoft and I don't want to meet their lawyers.

They have also another soft, based on Glicko http://smashboards.com/threads/glickoman-play...er.328804/

Why dont ask MS for permission to use it? After all u don't do it fot commersial thing/making money of it. Maybe they will let u use it for free. :)

They might if you gave them some advertising space. Worth asking if it's worth using. Sounds like you're not convinced it's worth using though either.

I ran the data through trueskill too. Several different implementations.
Not a single javascript implementation I found produced the same data as the official Microsoft online Trueskill calculator for a single match.

The python implementation was consistent with Microsoft's results, by unfortunately I don't speak python. And the code I had to test it was too slow to allow many test runs.

Looking at the top 20 of the glicko and Trueskill lists returned the same names, some in slightly different order.

I can't help you with the coding part but I'm graduate in math: if you need help, don't hesitate to write me or add me on steam, I'll be happy to answer.

pure raw score <3

fuck k/d ratio [ and with it ffa/duel playstyle @ CA ]

too early for judge

Am i the only guy who thought the benefits outweighed the negatives regarding rankings?

no, it's just that minorities are usually much more outspoken, than the majority

Love how guys regurgitate the same old silly crap about elo . If you never read anything about it, you dont know the difference between implementations, you just "play the game for fun" just shut your trap. You opinions are just as valid as 12 year old boys who hates math. Math rocks, probability rules the universe, elo shuffling is better than doing nothing and your opinion on personal grounds means jackshit nothing at all. Thank you for listening.

it would be best if elo was calculated only by individual perfomance aka damage dealt/score per time played(if joined running match) etc. unlike old qlranks when u could elo boost just joining winning team.
I dont know how this things can be/are done but so far i can tell this system is much better than the old one so thx and gl

if i understood corectly predathor reply (post #6) he is going for individual calculation.

is exactly how new elo works

People still caring about stats. Get well soon.

bye_world

Well, I think that, in team games, whether your team wins or not SHOULD be taken into consideration. I've already noticed that players are playing for personal score and don't care about the result on servers running the system. Sure, people used to cry about how they lose all their elo because of bad teammates, but you have the same chance of the bad players being on the enemy team. In fact...you have MORE chance that the bad players will be on the enemy team, because there are 4 of them. (in a 4v4) There are only 3 possible positions on your own team that could be filled by a bad player, since you're not one yourself, right...?

Mostly, I just like the accuracy graph. It shows some interesting patterns. Plus, if you want to make a concerted effort to improve with a weapon, it's handy to have a graph to look at. That's the kind of shit that scientists do: looking at graphs and wearing glasses and shit

i like that shit, graphs and shit, maybe some pies aswells.

There's pros and cons to both ways. When you can't enforce balanced teams (that is, players may decide in which team they play), then a system based on team win/loss is toxic.

Nobody wants to join a 4vs3 match when the smaller team is behind. Kick+F1 is so automated that nobody even reads the votes anymore. A player falls a little behind, kick and better guy on my team plz.

Yeah, matchmaking wouldn't be too hard to do if not for all the little workarounds and dodges people use. There should be more options for server hosts to set strict automated functions like only allowing the first spec who joined to join the game, banning people who quit before the end of the game, etc. It's impossible for a rating system to be entirely accurate when people do all they can to artificially inflate/deflate their own rating.

Still, keep up the good work. It's always gonna be better than a rando shuffle once there's a enough data. Currently, I'm seeing a lot of 1000 elo guys destroying people, but that's to be expected since it's the holidays and more people are finding time to play.

glicko <3

\,,/ o_X \,,/


happy xmas ESR :)