OakEmu-RL v0.1.0

Appearance

PPOModelConfiguration ​

Configuration for the PPO actor-critic model architecture.

Source

Values ​

conv_activation ​

python
conv_activation: ActivationName = 'relu'

Activation function for convolutional layers.

Source

conv_filters ​

python
conv_filters: list[tuple[int, int | tuple[int, int], int | tuple[int, int]]] = [(32, 8, 4), (64, 4, 2), (64, 3, 1)]

Convolutional filters: (out_channels, kernel_size, stride).

Source

fcnet_activation ​

python
fcnet_activation: ActivationName = 'relu'

Activation function for fully connected layers.

relu units that are pushed negative for every input switch off for good (zero output, zero gradient); over a long run that shrinks the layer. leaky_relu, elu, gelu and silu keep a gradient on the negative side, so a unit can come back.

Source

fcnet_hiddens ​

python
fcnet_hiddens: list[Annotated[int, Field(gt=0)]] = [512]

Hidden layer sizes for the shared fully connected feature extractor.

Source

orthogonal_init ​

python
orthogonal_init: bool = True

Whether to use orthogonal weight initialization.

Source