Implementation of the TrueSkill algorithm

Description

An R implementation of the TrueSkill Algorithm (Herbrich, R., Minka, T. and Grapel, T. [1]), a Bayesian skill rating system with inference by approximate message passing on a factor graph. Used by Xbox to rank gamers and identify appropriate matches.

http://research.microsoft.com/en-us/projects/trueskill/default.aspx

Current version allows for one player per team. Will update as time permits. Requires R version 3.0 as it is implemented with Reference Classes.

The code for the examples can be found at:

system.file('', package = 'trueskill')

Acknowledgements to Doug Zongker [2] and Heungsub Lee [3] for their python implementations of the algorithm and for the liberal reuse of Doug's code comments.

Details

Main Functions and Classes

Reference Classes:

Gaussian, Player, Parameters

Methods:

Multiply, Divide

Functions:

AdjustPlayers, Trueskill, DrawMargin, DrawProbability, PrintList

Data:

data

Author(s)

Brendan Houng <brendan.houng@gmail.com>

References

[1]

TrueSkill: A Bayesian Skill Rating System, Herbrich, R., Minka, T. and Grapel, T.

[2]

Doug Zongker's python implementation:
https://github.com/dougz/trueskill

[3]

Heungsub Lee's python implementation:
https://github.com/sublee/trueskill.

[4]

Jeff Moser's explanatory notes:
http://www.moserware.com/2010/03/computing-your-skill.html

Examples

  # Example 1.
  
  # set default values for BETA, EPSILON and GAMMA where BETA is sigma / 2
  # EPSILON is DrawProbability(0.1)
  # GAMMA is sigma / 100
  parameters <- Parameters$new()
  
  Alice  <- Player(rank = 1, skill = Gaussian(mu = 25, sigma = 25 / 3), name = "1")
  Bob    <- Player(rank = 2, skill = Gaussian(mu = 25, sigma = 25 / 3), name = "2")
  Chris  <- Player(rank = 2, skill = Gaussian(mu = 25, sigma = 25 / 3), name = "3")
  Darren <- Player(rank = 4, skill = Gaussian(mu = 25, sigma = 25 / 3), name = "4") 
   
  players <- list(Alice, Bob, Chris, Darren)
  
  players <- AdjustPlayers(players, parameters)  
  PrintList(players)
  print(Alice$skill)

  # Relying on positional arguments looks much cleaner:
  Alice  <- Player(1, Gaussian(25, 8.3), "Alice")
  Bob    <- Player(2, Gaussian(25, 8.3), "Bob")
  Chris  <- Player(2, Gaussian(25, 8.3), "Chris")
  Darren <- Player(4, Gaussian(25, 8.3), "Darren") 
 
  # Example 2 - see https://gist.github.com/bhoung/5596282  
  # the example applies trueskill to tennis tournament data
  # (runtime is approx 50 secs)
  

Results