Source code for openskill.models.thurstone_mosteller_full

import math
from typing import Callable

from openskill.constants import Constants
from openskill.statistics import v, vt, w, wt
from openskill.util import gamma


[docs]class ThurstoneMostellerFull: def __init__(self, game, team_rating: Callable, **options): self.constants = Constants(**options) self.EPSILON = self.constants.EPSILON self.TWO_BETA_SQUARED = self.constants.TWO_BETA_SQUARED self.team_ratings = team_rating(game, **options) self.gamma = gamma(**options)
[docs] def calculate(self): result = [] for i, i_team_ratings in enumerate(self.team_ratings): omega = 0 delta = 0 i_mu, i_sigma_squared, i_team, i_rank = i_team_ratings for q, q_team_ratings in enumerate(self.team_ratings): q_mu, q_sigma_squared, q_team, q_rank = q_team_ratings if q == i: continue ciq = math.sqrt( i_sigma_squared + q_sigma_squared + self.TWO_BETA_SQUARED ) delta_mu = (i_mu - q_mu) / ciq sigma_squared_to_ciq = i_sigma_squared / ciq gamma = self.gamma(ciq, len(self.team_ratings), *i_team_ratings) if q_rank > i_rank: omega += sigma_squared_to_ciq * v(delta_mu, self.EPSILON / ciq) delta += ( gamma * sigma_squared_to_ciq / ciq * w(delta_mu, self.EPSILON / ciq) ) elif q_rank < i_rank: omega += -sigma_squared_to_ciq * v(-delta_mu, self.EPSILON / ciq) delta += ( gamma * sigma_squared_to_ciq / ciq * w(-delta_mu, self.EPSILON / ciq) ) else: omega += sigma_squared_to_ciq * vt(delta_mu, self.EPSILON / ciq) delta += ( gamma * sigma_squared_to_ciq / ciq * wt(delta_mu, self.EPSILON / ciq) ) intermediate_result_per_team = [] for j, j_players in enumerate(i_team): mu = j_players.mu sigma = j_players.sigma mu += (sigma**2 / i_sigma_squared) * omega sigma *= math.sqrt( max(1 - (sigma**2 / i_sigma_squared) * delta, self.EPSILON), ) intermediate_result_per_team.append([mu, sigma]) result.append(intermediate_result_per_team) return result