A code-base for 'Explaining RL Decisions with Trajectories' published at ICLR 2023:

Quick Blog | Full Paper

---

Here we provide the code for the gridworld experiments, which can be found in the gridworld_expts.ipynb file. We hope the implementation for other environments would be clear from this example. In case of additional queries, feel free to reach out at: [email protected]

---

Instructions for usage:

1. Before running the code-base, install the dependencies using:

       conda create -n xrl python=3.8 -y
       conda activate xrl
       pip install -r requirements.txt
       python -m ipykernel install --user --name xrl
   

2. Launch gridworld_expts.ipynb using a jupyter server. Activate the xrl kernel and run the file to generate the results from the paper.

Acknowledgements: We use Dynamic Programming implementation from andrecianflone/dynaq/ and we are thankful to the authors for making it publicly available.

Citation

If you use this code for your research, please cite our paper:

@misc{deshmukh2023explaining,
      title={Explaining RL Decisions with Trajectories}, 
      author={Shripad Vilasrao Deshmukh and Arpan Dasgupta and Balaji Krishnamurthy and Nan Jiang and Chirag Agarwal and Georgios Theocharous and Jayakumar Subramanian},
      year={2023},
      eprint={2305.04073},
      archivePrefix={arXiv},
      primaryClass={cs.AI}
}