torp

TORP ratings, match predictions and season simulations built with this package are published, updated after every round, at inthegame.blog/afl.

torp is an R package for AFL analytics. It provides player ratings (TORP), expected points and win probability models, season simulations, and easy access to processed AFL data.

Installation

# install.packages("devtools")
devtools::install_github("peteowen1/torp")

# For the latest pre-trained models
devtools::install_github("peteowen1/torpmodels")

Quick Start

Load Data

All data is stored as parquet files on torpdata GitHub releases and loaded on demand.

library(torp)

# Play-by-play data
pbp <- load_pbp(2025, rounds = 1:10)

# Player statistics
player_stats <- load_player_stats(2025)

# Match results and fixtures
results <- load_results(2025)
fixtures <- load_fixtures(all = TRUE)

# All available data (2021+)
pbp_all <- load_pbp(TRUE, rounds = TRUE)

Player Ratings

TORP blends two independent rating systems: EPR (from play-by-play data) and PSR (from box-score stats via Bayesian estimation + glmnet). Per-game equivalents EPV and PSV measure single-game contribution.

# Career TORP ratings (EPR + PSR blend)
ratings <- torp_ratings()

# Per-game values (EPV + PSV + torp_value)
game_ratings <- player_game_ratings(season_val = 2026)

# Season totals
season_totals <- player_season_ratings(2026)

# Bayesian stat ratings (per-stat next-game predictions)
stat_ratings <- load_player_stat_ratings(2026)

Prediction Models

Three core models add expected points, win probability, and shot outcome predictions to play-by-play data.

pbp <- load_pbp(2025, rounds = 1:5)
pbp_clean <- clean_pbp(pbp)

# Add model predictions
pbp_ep <- add_epv_vars(pbp_clean)     # Expected points
pbp_wp <- add_wp_vars(pbp_ep)         # Win probability

Season Simulation

# Simulate remaining matches based on TORP ratings
sim_result <- simulate_season(sim_teams, sim_games)

Package Ecosystem

Package	Role	Links
torp	Core analytics, data loading, ratings, models	GitHub
torpdata	Processed AFL data via GitHub releases (parquet)	GitHub
torpmodels	Pre-trained models via GitHub releases (RDS)	GitHub

Data flows from the AFL API (via torp’s get_afl_*() functions) through torp’s cleaning pipeline into torpdata, with models stored in torpmodels.

Learn More

• vignette("getting-started") – Installation and data loading
• vignette("torp-guide") – Ratings, models, data architecture, and simulation

License

MIT