TrueSkill Ranking System (in the wild)
So I’m in a poker league. We play three out of four Thursday nights in a month for a total of fifteen events in a season. Each event you earn a certain number of points. You’re 10 best finishes based on points out of the fifteen events are counted. After the grueling fifteen week season, there is a finale where you start with an amount of chips proportional to the number of points you earned during the season. Anyway, another player (Shaun) and I took over the scoring for the league this season. Our scoring system is pretty basic (50 points for showing up, 50 points for everyone you beat, 600-300-150-75 bonus for cashing (finishing top 4)). On top of this I’ve devised a fairly reasonable ranking system separate from the points based on your average finish and how many events you have played. One criticism of my ranking system is that I’m not accounting for the strength of the field, I’m just looking at average percentile of finish.
So I usually car pool to and from events with Shaun and we’ve been talking about ranking systems. Last night he mentioned that he had been doing some research on how Xbox live does their rankings. Each player has some level of ability and an uncertainty associated with their skill level. After each game players rankings, skill level and uncertainty, are updated. He described it a little bit more, and I mentioned that it sounded Bayesian to me. Turns out it is!
Here is an introduction to the TureSkill ranking system and here is a more detailed description. For those of you who are interested in all the details, here is the paper “TrueSkill(TM): A Bayesian Skill Rating System” where they propose the system.
P.P.S. I finished second in the regular season last year, but won the finale. This year I was briefly in first place until last night. I am now in second place by 150 points with 4 events to play.