Models

openskill ships five rating models. They all live in the openskill/models subpath.

import {
  plackettLuce,
  bradleyTerryFull,
  bradleyTerryPart,
  thurstoneMostellerFull,
  thurstoneMostellerPart,
} from 'openskill/models'

Pass one to rate via the model option:

import { rate } from 'openskill'
import { bradleyTerryFull } from 'openskill/models'

const [[a2], [b2]] = rate([[a1], [b1]], { model: bradleyTerryFull })

Plackett-Luce — default

A generalized Bradley-Terry model that scales to k ≥ 3 teams without needing a k−1 dimensional integration. Use this unless you have a reason not to.

Bradley-Terry

Logistic distribution over player skill — similar to Glicko.

• bradleyTerryFull — full pairing. More accurate, but every team is compared against every other team.
• bradleyTerryPart — partial pairing. Each team is compared only with its ranking neighbors: the adjacent teams, or the teams at most windowSize places away when you set it. Use this for high-k matches (e.g. 100-player races) where a full pairing would be expensive.

Thurstone-Mosteller

Gaussian distribution over player skill — similar to TrueSkill. Accuracy is generally a touch worse than Bradley-Terry, but it can be tuned.

• thurstoneMostellerFull — full pairing.
• thurstoneMostellerPart — partial pairing for high-k matches, with the same windowSize.

Choosing a model