Here, we present Glicko2 Duel Predictions for the 22 active players we have currently on the rooster for QL Duel. We have 2.1 million matches in the database, 75 weeks of stats calculations.

As always, the point marks the current ranking power on that map, the bars width marks the 95% confidence interval we have for that player on said map.

Players are ordered on the geometric mean of the 5 maps. This means the prediction is that rapha is currently the strongest player, and so on. Do note that if the seeding is bad, then the final results might be very different.

Also, it looks like Cure is the interesting map which can set up everything :)

Also, PDF is downloadable:

https://dl.dropboxusercontent.com/u/196031/ma...on2013.pdf

There will probably be a post which is more up-to-date later next week when we know more about the player list.

I've been pimping this before, but I'll create another thread and do it again with updated data. Here is the current matchup data:

Dream Hack 2012 Winter Matchups

The ordering is based on a conservative estimate of the players skill given by the mean of the lower end of the rating deviance. That is, the order in which players occur is their relative strength, topmost being the strongest. Note that strong players have no maps on which they are weak due to having a good mean. I tried the median as well which makes the top players: k1llsen, rapha, cypher, evil in that order.

The point is the current rating of the player. The bar is the variance of the rating. That is, the system has 95% confidence that the player is in this range of rating. Thus, if there is a lot of overlap, it can be either players game.

Update: Now we use the geometric mean over the arithmetic mean.

So, I've been working more on the ranking system. It is currently tracking 1.12 million matches or so, around half of those on Blood Run/Ztn :)

This time around, I'll just link to the PDF documents which contains the rankings. There is one for each map in the DHW 2012 pool (I had to start somewhere, and I am out of Guinness). As always, the dot signifies the rating of the player and the bar signifies the confidence we have in the rating of a player. The ladders uses a conservative estimate and rewards players who plays more on the given map. Also, if one doesn't play for a while the rank on the ladder will slowly fall down.

Ranking is the Glicko 2 ranking system.

Currently, I am working on a banned-player remover because some of the players that snuck themselves into the list are obvious cheaters - if you ask quake live nicely, they are banned.

hektik:
Rankings
Ladder

aerowalk:
Rankings
Ladder

bloodrun:
Rankings
Ladder

silence:
Rankings
Ladder

lostworld:
Rankings
Ladder

furiousheights:
Rankings
Ladder

toxicity:
Rankings
Ladder

have fun!

So with a new LAN coming up, here are the rankings for a pretty recent run.

I predict, based upon my 1.1 million matches that k1llsen wins, by beating evil. Third place goes to pavel because he beats baksteen.

You may claim that this is wrong, but based on their online performance recently, this is the outcome of the tournament.

Here is a kernel density plot of duelers on the 5 "main" maps: t7, dm13, tox, aero/hub and ztn. The x axis is the ranking of a player in Glicko2, the y is the density.

There are two interesting developments:

First, I am now ranking players individually on maps. So I know that I, Relativizor is miserable on hub and my best map is dm6 at the moment with ztn a close second. This was started because of something fazz said and I started to wonder about it.

I still don't know very much how I should draw the rankings, but I'll probably just draw it individually for each map.

Second, if you take a look at the above graph, there is a concentration of players around 1700 Glicko (I guess this is around 1500 ELO). This means that there are many players around that rank so it is technically where most of the players are. I don't count the large group who just played duel, lost and never came back. In any case, I found the data interesting. So may you.

Note: a1r-r41d has not played for 9 months, so he is out!

The above lists the current ratings of the invitees we have up until now (19 players). They are rated by Glicko2 based on matches from around february and up until now. The middle "dot" is where we estimate the players strength is. The width of the horizontal line gives a 95% confidence interval of the rating.

For instance, Cooller and rapha have roughly the same strength, but since rapha plays way more than Cooller, we have more confidence in rapha's skill than Cooller's. Or said in another way: We know absolutely nothing of Cooller's strength. He may be way better than rapha or worse than, say, twister.

p0ni is very blue. This means that he has a high "stochastic volatility" in the sense that he is beating players who are better than him from time to time. Or said another way: p0ni is likely to upset and win games against the stronger opponents. Or lose against weaker ones.

As we get closer, I might update the graph, but I do know on whom I would bet my money as it stands right now :P

This image represents ratings from 4 tournaments in 2011: IEM5, DreamHack Summer 2011, QuakeCon 2011 and DreamHack Winter 2011. It rates all players present at those tournaments and plots their current rating according to the Glicko-2 Rating system.

Unlike ELO, G2 is much more advanced. First of all, we are not reporting a single value as the rating of a player. Rather, we are reporting an interval in which we have 95% confidence.

As a conservative estimate, we use the lowest point (marked as a black dot) as the current skill of the player. This penalizes players that are not playing. Also, the color chosen reflects a third value, namely how "volatile" the players rating is.

In Glicko 2, we have three numbers for each player: Rating - R, Variance - RD, and Volatility - V. The rating works much like in ELO. The higher the number, the better you are. But unlike ELO we also associate a variance with each rating, RD. As you play games your variance will shrink down. This means you will move around less once you play enough games. If you don't play for a period, your variance is increased. This reflects the systems confidence in your current skill rating.

Finally, we have the volatility, which reflects how erratic the players performance is. If you suddenly win lots of matches after a period of stability (or likewise lose them), the volatility will be high. Do note that all these players have fairly low volatility - it is just the measurement gradient scale that makes it seem exaggerated.