Automated Feature Selection for Inverse Reinforcement Learning

Overview

This repository contains the source code and data for the paper Automated Feature Selection for Inverse Reinforcement Learning

Installation

• Clone the repo git clone [email protected]:baimukashev/feature4irl.git
• Run cd feature4irl
• Run pip install -r requirements.txt
• Optionally, setup Weight and Biases for tracking the experiments.

How to run the code

• Run ./experiments_{acr, pend, cart}.sh for training all baselines for each task respectively.
• Results of the experiments including the simulation videos, data will be then saved in ./checkpoints folder.

Configuration

Configuration files for each environment are located in feature4irl/cfg/.

Citation

@misc{baimukashev2024automatedfeatureselectioninverse,
      title={Automated Feature Selection for Inverse Reinforcement Learning}, 
      author={Daulet Baimukashev and Gokhan Alcan and Ville Kyrki},
      year={2024},
      eprint={2403.15079},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2403.15079}, 
}