我有附加的模型(由RasmusBååth提供),我正在努力改进以增加我对MCMC和R的理解。美联储进入该模型的是足球比赛,每一行包含(HomeTeam,AwayTeam,Season,HomeGoals, AwayGoals,MatchResult,HomeForm,AwayForm)。该模型用于预测两个团队之间的游戏中的足球结果,其中HomeGoals和AwayGoals被建模为泊松分布。我过去几天试图完成的是使用关于模型中团队当前形式的数据,但我想不出实现它的方法。我希望能提出一些建议。
一些澄清:
谢谢!
MODEL:
model {
for(game_i in 1:n_games) {
HomeGoals[game_i] ~ dpois(lambda_home[Season[game_i], HomeTeam[game_i], AwayTeam[game_i]])
AwayGoals[game_i] ~ dpois(lambda_away[Season[game_i], HomeTeam[game_i], AwayTeam[game_i]])
}
for(season_i in 1:n_seasons) {
for(home_i in 1:n_teams) {
for(away_i in 1:n_teams) {
lambda_home[season_i, home_i, away_i] <- exp(home_baseline[season_i] + skill[season_i, home_i] - skill[season_i, away_i])
lambda_away[season_i, home_i, away_i] <- exp(away_baseline[season_i] + skill[season_i, away_i] - skill[season_i, home_i])
}
}
}
skill[1, 1] <- 0
for(j in 2:n_teams) {
skill[1, j] ~ dnorm(group_skill, group_tau)
}
group_skill ~ dnorm(0, 0.0625)
group_tau <- 1/pow(group_sigma, 2)
group_sigma ~ dunif(0, 3)
home_baseline[1] ~ dnorm(0, 0.0625)
away_baseline[1] ~ dnorm(0, 0.0625)
for(season_i in 2:n_seasons) {
skill[season_i, 1] <- 0
for(j in 2:n_teams) {
skill[season_i, j] ~ dnorm(skill[season_i - 1, j], season_tau)
}
home_baseline[season_i] ~ dnorm(home_baseline[season_i - 1], season_tau)
away_baseline[season_i] ~ dnorm(away_baseline[season_i - 1], season_tau)
}
season_tau <- 1/pow(season_sigma, 2)
season_sigma ~ dunif(0, 3)
}
DATA(部分内容):
No. HomeTeam AwayTeam Season HomeGoals AwayGoals HomeForm AwayForm MatchResult
41 Betis Real Madrid 0809 1 2 -0.6 0.6 -1
42 Espanol Barcelona 0809 1 2 0.0 0.4 -1
43 Sp Gijon Villarreal 0809 0 1 -1.0 0.8 -1
44 Almeria Recreativo 0809 1 0 0.4 -0.4 1
45 Ath Bilbao Getafe 0809 0 1 -0.2 0.2 -1
46 Ath Madrid Sevilla 0809 0 1 0.2 0.6 -1
47 Malaga Valladolid 0809 2 1 -0.4 -0.2 1
48 Numancia Osasuna 0809 0 0 -0.4 -0.2 0
49 Santander Mallorca 0809 1 2 -0.6 0.2 -1
50 Valencia La Coruna 0809 4 2 0.8 -0.2 1
51 Barcelona Ath Madrid 0809 6 1 0.8 -0.2 1
52 Villarreal Betis 0809 2 1 1.0 -0.6 1