Run Training

Launch basic training with command: python cg_runner.py --policy <policy> --env <env>. Progress logging and checkpoints can be found in the /save directory.

For other argument options, look into args.py, env_maker.py and policy_maker.py.

Currently available <policy>:

  • de (decentralized)
  • dicg_ce (DICG-CE)
  • proximal_cg (proximity-based coordination graph)

Currently available <env>:

  • meet (meeting in the grid world)
  • predprey (predator-prey)
  • traffic (hard mode traffic junction)