FRLlibSACSettings

A struct to hold Soft Actor-Critic (SAC) settings for an RLlib training script.

struct FRLlibSACSettings : public FTrainingSettings

Methods

GenerateTrainingArgs

virtual void GenerateTrainingArgs(FScriptArgBuilder &ArgBuilder) const

Generate the training arguments for the script using the ArgBuilder.

Parameters

• ArgBuilder (FScriptArgBuilder)

---

FRLlibSACSettings

virtual ~FRLlibSACSettings()

Attributes

Tau

float Tau = 0.005f

Soft update coefficient for target networks (0.0-1.0).

---

TEXT

FString TargetEntropy = TEXT("auto")

Target entropy for automatic temperature tuning.

Parameters

• "auto"

---

InitialAlpha

float InitialAlpha = 1.0f

Initial temperature/alpha for entropy regularization.

---

NStep

int NStep = 1

Number of steps for n-step returns.

---

bTwinQ

bool bTwinQ = true

Whether to use twin Q networks (double Q-learning) to reduce overestimation bias.

Source: Source/ScholaTraining/Public/TrainingSettings/Ray/Algorithms/RLlibSACSettings.h