Skip to content

RolloutBufferTrainableDiagnostics ​

python
RolloutBufferTrainableDiagnostics(values: np.ndarray, returns: np.ndarray, advantages: np.ndarray) -> None

Flattened trainable rollout statistics used for learner-side diagnostics.

Source

Values ​

advantages ​

python
advantages: np.ndarray

Flattened policy advantages for trainable rollout items.

Source

returns ​

python
returns: np.ndarray

Flattened returns for trainable rollout items.

Source

values ​

python
values: np.ndarray

Flattened value predictions for trainable rollout items.

Source