Supporting code for the paper DERAIL: Diagnostic Environments for Reward and Imitation Learning.

The environments are available at the HumanCompatibleAI/seals repo. This repo contains the rest of the code for running the experiments in the paper.

To reproduce the results:

git clone https://github.com/HumanCompatibleAI/derail
cd derail
pip install .

python -m derail.run -t 500000 -n 15 -p
python -m derail.plot -f results/last.csv