OakEmu-RL v0.1.0

Appearance

RolloutMinibatch ​

python
RolloutMinibatch(observations: torch.Tensor, actions: torch.Tensor, log_probs: torch.Tensor, values: torch.Tensor, advantages: torch.Tensor, returns: torch.Tensor) -> None

A minibatch of rollout data for PPO training.

Source

Values ​

actions ​

python
actions: torch.Tensor

Actions with shape (minibatch_size,).

Source

advantages ​

python
advantages: torch.Tensor

GAE advantages, shape (minibatch_size,).

Source

log_probs ​

python
log_probs: torch.Tensor

Log probabilities of the actions under the old policy, shape (minibatch_size,).

Source

observations ​

python
observations: torch.Tensor

Observations with shape (minibatch_size, *obs_shape).

Source

returns ​

python
returns: torch.Tensor

Discounted returns (advantages + values), shape (minibatch_size,).

Source

values ​

python
values: torch.Tensor

Value estimates from the old policy, shape (minibatch_size,).

Source