OakEmu-RL v0.1.0

Appearance

PPORolloutMetrics ​

python
PPORolloutMetrics(policy_loss: float = 0.0, value_loss: float = 0.0, behavior_cloning_loss: float = 0.0, entropy: float = 0.0, loss: float = 0.0, approx_kl: float = 0.0, clip_fraction: float = 0.0, ratio_mean: float = 0.0, grad_norm: float = 0.0, num_updates: int = 0) -> None

Accumulated learner metrics across one full rollout update.

Source

Methods ​

add ​

python
add(minibatch: PPOMinibatchMetrics) -> None

Accumulate one minibatch's learner metrics.

Parameters:

  • minibatch (PPOMinibatchMetrics) – The minibatch metrics to accumulate.

Source

average ​

python
average(value: float) -> float

Return the rollout-average for an accumulated metric.

Parameters:

  • value (float) – The accumulated metric total.

Returns:

  • float – The average value across all updates.

Source

Values ​

approx_kl ​

python
approx_kl: float = 0.0

Source

behavior_cloning_loss ​

python
behavior_cloning_loss: float = 0.0

Source

clip_fraction ​

python
clip_fraction: float = 0.0

Source

entropy ​

python
entropy: float = 0.0

Source

grad_norm ​

python
grad_norm: float = 0.0

Source

loss ​

python
loss: float = 0.0

Source

num_updates ​

python
num_updates: int = 0

Source

policy_loss ​

python
policy_loss: float = 0.0

Source

ratio_mean ​

python
ratio_mean: float = 0.0

Source

value_loss ​

python
value_loss: float = 0.0

Source