Multi-Objective Model-based Reinforcement Learning for Infectious Disease Control

This repository is the official implementation of the paper "Multi-Objective Model-based Reinforcement Learning for Infectious Disease Control".

Requirements

• R version 3.6.3 (2020-02-29)

• Main packages for the proposed estimation procedure

  • abind
  • crayon
  • doFuture
  • doParallel
  • foreach
  • ggplot2
  • ggpubr
  • grid
  • gridExtra
  • matrixStats
  • parallel
  • plyr
  • randomForest

• Additional packages for experiments

  • gtools
  • dplyr
  • tidyr

File Overview

1. Files in the main folder: scripts for reproducing results.

• sir_pred.R: script to conduct estimation and validation of the transition model.
• pareto_eval.R: script to conduct evaluation of the Pareto optimal policies.
• simu_H1N1: script to conduct evaluation of the Pareto optimal policies for the H1N1 experiment.
• simu_valid.R: script to conduct cross validation of the environment model.
• robustness.R: script to conduct the hyper-parameter sensitivity analysis.
• data.rds: the data file, consisting of action, observations, and costs of the six cities.

2. Files in the /code folder

   1. the proposed method
      • SIR.R: main functions for the generalized dynamics model
      • RL.R: main functions for the reinforcement learning algorithms
      • utility.R: helper functions for SIR model and RL parts
   2. experiments
      • simu.R: main functions for simulation experiments
      • simu_utility.R: helper functions for simulation experiments

3. Files in the /SEIR folder: script to conduct evaluation of the Pareto optimal policies when the DGP is the SEIR model.

Reproduce Results

To reproduce ourexperiment results in the paper, open the corresponding script, change the working directory to the main folder which includs this README file, modify the main_path in the script, and run commands below.