Inverse Reinforcement Learning in Contextual MDPs

This repository contains the code files that we used in our work to construct three environments and test three learning methods. as well as comparing to baselines. The environments in this repository:

1. Autonomous driving simulator
2. Dynamic treatment regime
3. Grid world

How to run the simulations

To recreate our simulations firstly clone this git and then run the scripts specified below, each one from its own path.

Plot the results

• use the jupyter notebooks in each environment.

Data required

in this work we use the processed data from point85AI git repository that can be found at:
https://github.com/point85AI/Policy-Iteration-AI-Clinician

The data set we use to construct the dynamic treatment regime can be found at:

• Policy-iteration-AI-Clinician/data/normalized_data.mat

It should be placed at:

• data/dynamic_treatment/all/normalized_data.mat