TrueSkill, the video game rating system
Julia translation of the Python TrueSkill project.
This TrueSkill project is opened under the BSD license but the TrueSkill(TM) brand is not. Microsoft permits only Xbox Live games or non-commercial projects to use TrueSkill(TM). If your project is commercial, you should find another rating system.
TrueSkill is a rating system among game players. It was developed by Microsoft Research and has been used on Xbox LIVE for ranking and matchmaking service. This system quantifies players’ TRUE skill points by the Bayesian inference algorithm. It also works well with any type of match rule including N:N team game or free-for-all.
This project is a Julia package which implements the TrueSkill rating system.
using Pkg
pkg"add https://github.com/kose-y/TrueSkill.jl"In TrueSkill, rating is a Gaussian distribution which starts from N(25,2532). μ is an average skill of player, and σ is a confidence of the guessed rating. A real skill of player is between μ±2σ with 95% confidence.
using TrueSkill
Rating()Rating(mu=25.0, sigma=8.333333333333334)
If some player’s rating is higher β than another player’s, the player may have about a 76% chance to beat the other player. The default value of β is 25/6.
Ratings will approach real skills through few times of the TrueSkill's Bayesian inference algorithm. How many matches TrueSkill needs to estimate real skills? It depends on the game rule. See the below table:
| Rule | Matches |
|---|---|
| 16P free-for-all | 3 |
| 8P free-for-all | 3 |
| 4P free-for-all | 5 |
| 2P free-for-all | 12 |
| 2:2:2:2 | 10 |
| 4:4:4:4 | 20 |
| 4:4 | 46 |
| 8:8 | 91 |
r1 = Rating()
r2 = Rating()
new_r1, new_r2 = rate(r1, r2)(Rating(mu=29.39583169299151, sigma=7.17147580700922), Rating(mu=20.604168307008482, sigma=7.17147580700922))
User should put the winner first. After the game, TrueSkill recalculates their ratings by the game rseult. You can also handle a tie game:
r1 = Rating()
r2 = Rating()
new_r1, new_r2 = rate(r1, r2; drawn=true)(Rating(mu=24.999999999999993, sigma=6.457515683245051), Rating(mu=24.999999999999993, sigma=6.457515683245051))
4-player free-for-all:
r1 = Rating()
r2 = Rating()
r3 = Rating()
r4 = Rating()
new_r1, new_r2, new_r3, new_r4 = rate([r1, r2, r3, r4])4-element Array{Rating,1}:
Rating(mu=33.20668089498382, sigma=6.348109386291753)
Rating(mu=27.40145515734498, sigma=5.7871628096649355)
Rating(mu=22.598544842686724, sigma=5.787162809661523)
Rating(mu=16.793319105008695, sigma=6.348109386298544)
4-player free-for-all with ties:
r1 = Rating()
r2 = Rating()
r3 = Rating()
r4 = Rating()
new_r1, new_r2, new_r3, new_r4 = rate([r1, r2, r3, r4]; ranks=[1, 2, 2, 4])4-element Array{Rating,1}:
Rating(mu=31.56397232349745, sigma=6.4047036446727095)
Rating(mu=24.993092934523535, sigma=5.559362333660422)
Rating(mu=25.006907065417195, sigma=5.559362333660494)
Rating(mu=18.436027676561814, sigma=6.404703644728995)
2:2:2 match:
r1 = Rating()
r2 = Rating()
r3 = Rating()
r4 = Rating()
r5 = Rating()
r6 = Rating()
(r1, r2), (r3, r4), (r5, r6) = rate([[r1, r2], [r3, r4], [r5, r6]])3-element Array{Array{Rating,1},1}:
[Rating(mu=29.72018660358779, sigma=7.541668779519052), Rating(mu=29.720186603587795, sigma=7.541668779519052)]
[Rating(mu=25.000000000002768, sigma=7.34810769389046), Rating(mu=25.000000000002764, sigma=7.34810769389046)]
[Rating(mu=20.279813396409434, sigma=7.541668779519555), Rating(mu=20.279813396409434, sigma=7.541668779519555)]
2:1 match:
r1 = Rating()
r2 = Rating()
r3 = Rating()
t1 = [r1]
t2 = [r2, r3]
t1, t2 = rate([t1, t2]) # t1 defeats t22-element Array{Array{Rating,1},1}:
[Rating(mu=33.73067114899642, sigma=7.317365362867211)]
[Rating(mu=16.269328851003575, sigma=7.317365362867211), Rating(mu=16.269328851003575, sigma=7.317365362867211)]
The core ideas used in this project were described in "TrueSkill (TM): A Bayesian Skill Rating System" available at http://research.microsoft.com/apps/pubs/default.aspx?id=67956
Some concepts were based on Jeff Moser's code and documents, available at http://www.moserware.com/2010/03/computing-your-skill.html