trpo_mpi

trpo_mpi

• Original paper: https://arxiv.org/abs/1502.05477
• Baselines blog post https://blog.openai.com/openai-baselines-ppo/
• mpirun -np 16 python -m baselines.trpo_mpi.run_atari runs the algorithm for 40M frames = 10M timesteps on an Atari game. See help (-h) for more options.
• python -m baselines.trpo_mpi.run_mujoco runs the algorithm for 1M timesteps on a Mujoco environment.