OakEmu-RL v0.1.0

Appearance

MinibatchLossTensors ​

python
MinibatchLossTensors(policy_loss: torch.Tensor, value_loss: torch.Tensor, behavior_cloning_loss: torch.Tensor, entropy: torch.Tensor, loss: torch.Tensor, approx_kl: torch.Tensor, clip_fraction: torch.Tensor, ratio_mean: torch.Tensor) -> None

Loss and diagnostic tensors computed by one minibatch forward pass.

Field names match PPOMinibatchMetrics so metric values can be collected by name.

Source

Values ​

approx_kl ​

python
approx_kl: torch.Tensor

Source

behavior_cloning_loss ​

python
behavior_cloning_loss: torch.Tensor

Source

clip_fraction ​

python
clip_fraction: torch.Tensor

Source

entropy ​

python
entropy: torch.Tensor

Source

loss ​

python
loss: torch.Tensor

Source

policy_loss ​

python
policy_loss: torch.Tensor

Source

ratio_mean ​

python
ratio_mean: torch.Tensor

Source

value_loss ​

python
value_loss: torch.Tensor

Source