Bayesian Elo

For the implementation of Bayesian Elo as described in the blog post(s) , refer to src/bayesian_elo_parameter_estimation.R

You would need ProjectTemplate installed to easily load data sets and source functions. Once the package is installed run - library(ProjectTemplate) - load.project() to set up the environment.

Steps I followed during running the model(s) on a weekly basis during the 2018 season

Main data sources

• AFL tables for the past results https://afltables.com/afl/afl_index.html
• current schedule from https://fixturedownload.com
• Expert tipster data from Herald Sun [No longer used anymore]

Updating the model data weekly

• Download the csv file from https://fixturedownload.com and copy it to the data folder. Filename looks like afl.2018.AUSEasternStandardTime
• As of lately herald sun expert tips data is not available publicly so the following is not applicable.
• Get the herald sun expert tipsters data (google for herald sun expert tips with the round number)
  • Edit the download_expert_tips_run_script.R in src folder and add the new link and edit the round and link_str.
  • Run the script and make sure that the expert_tips variable looks OK. The script would save the results to an RDS file.
• Run clear.cache() and (re)run
  • library(ProjectTemplate)
  • load.project()

Run retro analysis for the previous rounds

• Run src/reto_score_weighted_expert_tips.R
• Run src/retro_score_simple_elo.R (make sure to update the round to the most recent completed round)
• Run the ensemble models
  • logistic: Run src/retro_score_ensemble_logistic.R after editing the round
  • Bayesian: Run src/retro_score_ensemble_bayesian.R after editing the round

Plot graphs to compare different models for the current season

• Run profiling/compare_models_historical_head_to_head.R

Run the predictions for the current round

• Run src/season_2018_predictions.R after editing the round