Implementation Code for DrilDICE (NeurIPS 2024)

• This code contains official implementation codes of DrilDICE and baselines for offline imitation learning.

1. Prerequisites

• To run this code, first install the anaconda virtual environment and install D4RL:

conda env create -f environment.yml
conda activate drildice
pip install d4rl

• (optional) Download D4RL dataset:

python download_d4rl_dataset.py

2. Generate Subsampling Indices for Covariate Shift Scenarios

python generate_subsample_indices.py

3. Train & Evaluate DrilDICE

• Train imitation policies using main.py.

Toy Domain:

• To run Four Rooms env., you should get MOSEK license.

pip install cvxpy[mosek] mosek

• Run cd fourrooms; python run_toy.py