Reinforcement Learning

We provide a full re-implementation of the PPO algorithm [1]. The policy supports three types of input features:

• obs: the observation from the base environment.

• hidden: the hidden representations of the novice.

• dist: the output softmax distribution of the novice.

Users can specify any non-empty combination of these feature types (e.g., obs_hidden, obs_hidden_dist).

References

[1]

John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. “Proximal Policy Optimization Algorithms.” arXiv preprint arXiv:1707.06347, 2017.