We'll see, according to that killsen or rapha should be the champ.

Problem is that graph doesnt measure lan performance only, so I would pick rapha/cypher, who are both top players and also usually great lan performers. Av3k, evil and killsen prolly completing top5.

a lan performance only graph would be neat, i got my money on cypher and some extra bucks on spart1e as usual

brilliant stuff

You said that you used all recent matches iirc? I don't see how it relates to dhw. I don't even see how it relates to random pub duel between two of these players.

With a ranking system like Glicko 2 or ELO your basic assumption is that the history of players performance can be used to predict their current performance. While having on-line games in the mix is not fair to LAN competition, it does yield a much larger corpora of data.

Quake Live has too few LAN tournaments so I can't really use those data. Worse, the tournaments are not played swiss style, but is single/double elimination with group play. This means you don't get to compare a large amount of players.

But even then, in the past this has often proven to be predictive of what happened.

It's not predictive. Everyone can tell you that cypher and rapha have higher chance of wining DHW than k1llsen.

The system actually predicts rapha to win over k1llsen. But I would also be a bit hesistant to call k1llsen as the 2nd place.

The situation is not different from rankings in e.g. Chess, where the player with the highest score of ELO is not necessarily the best player overall.

The situation is not different from rankings in e.g. Chess, where the player with the highest score of ELO is not necessarily the best player overall.

The fact that others are wrong too doesn't make it any better.
My point is that it clearly isn't predictive and it doesn't say any more about DHW than the obvious fact that top players are capable of beating each other. Intuition and reasoning provided with better quality data is far more accurate on small set of players.

Well, a Monkey randomly selecting matches has 50% prediction rate :P I hope that the system does predict way better than this. I think it was DHS 2012 where it accurately predicted top 4. Or perhaps that was Quakecon, can't remember.

At the latest GD Studio Arena it had a 88% prediction precision. Now, you may be able to do better with careful analysis, sure.

But there are other uses than prediction from such a system. Matchmaking is an obvious one. Another one, when we have per-map-rankings is to guage the relative strength of a given matchup on a given map. That is to me more interesting. For fun, how well does it predict map calls by the players?

gd arena was online
Ofc it's great for match making. I specifically wrote "small set of players" and "DHW".
It's great in predicting nicer vs rapha outcome but this is not what ppl want to know.
I don't think your method predicts dhw top3 any better than random esr user (excluding "1. cooller 2. cooller 3. cooller votes). There aren't even any arguments why it should do it better as it uses unrelated data.

What you do here is speculating using recent online results, but we've seen from the past that they don't correspond to lan performance.
I'd rather speculate using previous tournaments' outcome.

one day left \o/

ninja cat lololol

The ordering is based on a conservative estimate of the players skill given by the mean of the lower end of the rating deviance.

Could you elaborate on that? What's the difference between deviance and variance? Why is there more than one deviance per player and map?

Specifically, why is e.g. rapha above k1llsen on aerowalk even though he has a lower average and the same variance as k1llsen?

The lower end of the bar is called RDlo in the code. It is the conservative estimate of the players skill, which benefits players with more knowledge since they have played more on the map. I often use that number as a "ladder" rank of players skill rather than their rating value (the dot). Now, if we take all the RDlo's on all maps and take the mean of those, we get the average conservative estimate of the players rating.

k1llsen scores better than rapha on many maps, but on maps like silence and toxicity he pays the price compared to rapha. Thus rapha has the highest average score.

The other way is to take the median of the maps. I.e., successively removing the best and worst map of a player until there is only a single map left. That map becomes the players rating.

I should probably have used the geometric mean instead though since the arithmetic mean pays the price when you are performing badly on silence.

I keep the players in the same order because if I reorder them, the table is way harder to read in my opinion. You can't use spatiality anymore and there are too many colors for my suspect vision to tell them apart.

Two things I take out of this:

Out of the top 7 (rapha, cypher, killsen, avek, pavel, evil, spartie) everyone can realistically beat anyone on (almost) all of the maps.

On hektik and silence, the outcomes appear to be most unpredictable. I wonder whether that's only because of the low number of matches played (silence) or also because of the randomness of the map itself (hektik?).

Who cares about diagrams if Cypha wins anyway.

that is so awesome, how did you get the data for all the maps? I want them aswell! FOR ALL PALYERS IN THE WORLD :D

QLRanks data.

Good guess. But no. I have my own match scraper.