FRLlibTrainingSettings

Configuration struct for Ray RLlib training scripts.

Contains all settings necessary to launch and configure an RLlib training session, including algorithm selection, hyperparameters, logging, checkpointing, and resource allocation.

struct FRLlibTrainingSettings : public FTrainingSettings

Methods

GenerateTrainingArgs

virtual void GenerateTrainingArgs(FScriptArgBuilder &ArgBuilder) const

Generate command-line arguments for the RLlib training script.

Parameters

• ArgBuilder (FScriptArgBuilder)

---

FRLlibTrainingSettings

virtual ~FRLlibTrainingSettings()

Virtual destructor.

Attributes

Timesteps

int Timesteps = 8000

The total number of environment timesteps to train for.

---

LearningRate

float LearningRate = 0.0003

The learning rate for the optimizer.

---

MinibatchSize

int MinibatchSize = 128

The size of minibatches for gradient updates.

---

TrainBatchSizePerLearner

int TrainBatchSizePerLearner = 256

The training batch size per learner worker.

---

NumSGDIter

int NumSGDIter = 5

The number of SGD (stochastic gradient descent) iterations per training batch.

---

Gamma

float Gamma = 0.99

The discount factor (gamma) for future rewards.

---

LoggingSettings

FRLlibLoggingSettings LoggingSettings

The logging settings for the training script.

---

CheckpointSettings

FRLlibCheckpointSettings CheckpointSettings

The checkpoint settings for the training script.

---

ResumeSettings

FRLlibResumeSettings ResumeSettings

The resume settings for the training script.

---

NetworkArchitectureSettings

FRLlibNetworkArchSettings NetworkArchitectureSettings

The network architecture settings for the training script.

---

ResourceSettings

FRLlibResourceSettings ResourceSettings

The resource settings for the training script.

---

Algorithm

ERLlibTrainingAlgorithm Algorithm = ERLlibTrainingAlgorithm::PPO

The algorithm to use during training (e.g.

---

PPOSettings

FRLlibPPOSettings PPOSettings

PPO specific settings.

---

APPOSettings

FRLlibAPPOSettings APPOSettings

APPO specific settings.

---

IMPALASettings

FRLlibIMPALASettings IMPALASettings

IMPALA specific settings.

---

SACSettings

FRLlibSACSettings SACSettings

SAC specific settings.

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