bc

BC

We propose an implementation of the BC algorithm for offline RL using a MSE Loss

• bc.py is for continuous actions with MSE loss -- using d4rl
• bc_discrete.py is for dsicrete actions with cross-entropy -- using d4rl_atari

Note on openAI gym: The gym API is changing a lot... Don't hesitate to use pip install 'gym[all]==0.15.3' to ensure compatibility with D4RL