OakEmu-RL v0.1.0

Appearance

PPOMinibatchMetrics ​

python
PPOMinibatchMetrics(policy_loss: float, value_loss: float, behavior_cloning_loss: float, entropy: float, loss: float, approx_kl: float, clip_fraction: float, ratio_mean: float, grad_norm: float) -> None

Learner metrics emitted by one PPO/BC minibatch update.

Source

Values ​

approx_kl ​

python
approx_kl: float

Source

behavior_cloning_loss ​

python
behavior_cloning_loss: float

Source

clip_fraction ​

python
clip_fraction: float

Source

entropy ​

python
entropy: float

Source

grad_norm ​

python
grad_norm: float

Source

loss ​

python
loss: float

Source

policy_loss ​

python
policy_loss: float

Source

ratio_mean ​

python
ratio_mean: float

Source

value_loss ​

python
value_loss: float

Source