FRLlibNetworkArchSettings

A struct to hold network architecture settings for an RLlib training script.

struct FRLlibNetworkArchSettings : public FTrainingSettings

Methods

GenerateTrainingArgs

virtual void GenerateTrainingArgs(FScriptArgBuilder &ArgBuilder) const

Generate the training arguments for the script using the ArgBuilder.

Parameters

• ArgBuilder (FScriptArgBuilder)

---

FRLlibNetworkArchSettings

virtual ~FRLlibNetworkArchSettings()

Attributes

ActivationFunction

ERLlibActivationFunctionEnum ActivationFunction =
    ERLlibActivationFunctionEnum::ReLU

The activation function to use in the neural network.

---

FCNetHiddens

TArray<int> FCNetHiddens = {512, 512}

The number and width of hidden layers in the neural network.

---

MinibatchSize

int MinibatchSize = 256

The size of each minibatch.

---

bUseLSTM

bool bUseLSTM = false

Whether to use an LSTM layer in the model (for sequential/partial observability).

---

LSTMCellSize

int LSTMCellSize = 64

The size of the LSTM cell (number of units).

---

MaxSeqLen

int MaxSeqLen = 20

Maximum sequence length for stateful (LSTM) models.

Source: Source/ScholaTraining/Public/TrainingSettings/Ray/RLlibNetworkArchitectureSettings.h