saulabs / trueskill Goto Github PK

a ruby gem that implements the trueskill algorithm

Home Page: http://saulabs.net/projects/trueskill

License: MIT License

Ruby 100.00%

trueskill's Introduction

trueskill

trueskill is a rating-system for games with an arbitrary number of teams and players developed by Microsoft Research. It is based on the Glicko rating system and solves some major flaws of the ELO system.

Usage

Factor Graph

Example:

require 'rubygems'
require 'saulabs/trueskill'

include Saulabs::TrueSkill

# team 1 has just one player with a mean skill of 27.1, a skill-deviation of 2.13
# and an play activity of 100 %
team1 = [Rating.new(27.1, 2.13, 1.0)]

# team 2 has two players
team2 = [Rating.new(22.0, 0.98, 0.8), Rating.new(31.1, 5.33, 0.9)]

# team 1 finished first and team 2 second
graph = FactorGraph.new(team1 => 1, team2 => 2)

# update the Ratings
graph.update_skills

Score Based Bayesian Rating

As an extension of the basic TrueSkill algorithm, a score based Bayesian rating method is implemented. Similar to the standard approach, the actual skill is updated based on the outcome of the game. Instead of ranks, the score difference of playes determines the skill updates.

The approach implemented is a generalization of the Gaussian Score Difference model as proposed in Score-based Bayesian Skill Learning.

Example:

require 'rubygems'
require 'saulabs/trueskill'

include Saulabs::TrueSkill

# team 1 has just one player with a mean skill of 27.1, a skill-deviation of 2.13
# and an play activity of 100 %
team1 = [Rating.new(27.1, 2.13, 1.0)]

# team 2 has two players
team2 = [Rating.new(22.0, 0.98, 0.8), Rating.new(31.1, 5.33, 0.9)]

# team 1 wins by 10 points against team 2
graph = ScoreBasedBayesianRating.new(team1 => 10.0, team2 => -10.0)

# update the Ratings
graph.update_skills

Installation

To install the TrueSkill gem, simply run

[sudo] gem install trueskill

Add the following to your script:

require 'saulabs/trueskill'

Known issues

The calculation of the ranking probability is not yet implemented

Plans

Generalize the method from the team vs team case to a general case with arbitrary number of teams

Note on Patches/Pull Requests

Fork the project.
Make your feature addition or bug fix.
Add tests for it. This is important so I don't break it in a future version unintentionally.
Commit, do not mess with rakefile, version, or history. (if you want to have your own version, that is fine but bump version in a commit by itself I can ignore when I pull)
Send me a pull request. Bonus points for topic branches.

Copyright

See LICENSE for details.

trueskill's People

Contributors

Stargazers

Watchers

trueskill's Issues

the order of teams appears to change after 'update_skills'

Hi,

Thanks for going through the trouble of creating this code!

When I set team ranks in some arbitrary order, say, [3, 1, 2], I am not sure how to match the updated scores to the original teams. E.g. starting from equal mean assumptions, the above ordering of game ranks, leads to the first team on the list having the highest posterior mean.

I made a spec that is a reshuffle of one of your specs. I would think it should pass.

describe 'standard rating and reverse order of wins' do

before :each do
  TrueSkill::FactorGraph.new(@teams, [2,1]).update_skills
end

it "should change second player's rating to [29.395832, 7.1714755]" do
  @teams[1][0].should eql_rating(29.395832, 7.1714755)
end

it "should change first player's rating to [20.6041679, 7.1714755]" do
  @teams[0][0].should eql_rating(20.6041679, 7.1714755)
end

end

Thanks,
Mike.

update_skills can result in an infinite loop

Hi,

There is a bug in the class FactorGraph. For some inputs, the method update_skills goes into an infinite loop.

Here is a simple example to reproduce the problem

require 'saulabs/trueskill'
include Saulabs::TrueSkill

scores = [[45, 0.5, 0.6],[9, 0.9, 0.8],[54, 0.1, 0.7]]
graph = FactorGraph.new(scores, (1..scores.size).to_a)
graph.update_skills

Same problem happens with the triplets:

[52, 0.0, 0.9],[52, 0.9, 0.9],[86, 0.6, 0.9]
[9, 0.1, 0.6],[6, 0.0, 0.4],[47, 1.0, 0.3]
[63, 0.3, 0.5],[40, 0.9, 0.7],[70, 0.7, 0.7]
[95, 0.8, 0.1],[58, 0.0, 0.3],[11, 0.9, 0.8]
[83, 0.8, 0.8],[91, 0.7, 0.7],[86, 0.4, 1.0]
[17, 6, 0.1],[26, 9, 0.2],[74, 9, 0.3]
[15, 3, 0.1],[5, 2, 0.4],[53, 7, 0.5]
[40, 8, 0.4],[94, 5, 0.4],[88, 7, 0.9]
[43, 7, 0.5],[74, 3, 0.6],[87, 8, 0.5]
[26, 2, 0.6],[3, 3, 0.7],[64, 8, 0.9]
[4, 1, 0.5],[12, 1, 0.5],[39, 1, 0.5]
[10, 1, 1],[15, 1, 1],[67, 2, 0.5]

Thanks

graph = FactorGraph.new(team1 => 1, team2 => 2)
graph.update_skills

returns a different result than

graph = FactorGraph.new(team2 => 2, team1 => 1)
graph.update_skills

Publish a new gem with the latest changes

Please publish a new gem with the latest changes.

The documentation is updated and reciprocates the code.
But, the last published gem is the older one and does not have the new changes.