Strategic Advantage Weighting for Imitation Learning

Paper

Important Files

Experiment Files

• generate_data.py
  • Top level script used to simulate an agent in an environment
  • Serializes trajectory data to be used for training
• train.py
  • Top level script used to train a pytorch BC agent
  • See file for all available training options
• visualization.py
  • Top level script used to analyze the results of data generation/training
  • Superimposes heat map over gridworld to visualize metric of choice
  • See file for available options and uses
• play_gridworld.py
  • Spawns interactive GUI to allow user to interact with pre-defined gridworld environments

RL Files

• rl.py
  • All utils for running tabular RL on a gridworld environment
• deep_rl.py
  • code for training an on-policy, offline Q agent