1. Epic GamesDeveloper
  3. ULearningAgentsPolicy

ULearningAgentsPolicy

A policy that maps from observations to actions.

API > API/Plugins > API/Plugins/LearningAgents

A policy that maps from observations to actions.

   
Name ULearningAgentsPolicy
Type class
Header File /Engine/Plugins/Experimental/LearningAgents/Source/LearningAgents/Public/LearningAgentsPolicy.h
Include Path #include "LearningAgentsPolicy.h"

Syntax

UCLASS (MinimalAPI, BlueprintType, Blueprintable, Meta=(BlueprintSpawnableComponent))  
class ULearningAgentsPolicy : public ULearningAgentsManagerListener

Inheritance Hierarchy

Constructors

Name Remarks Include Path Unreal Specifiers
These constructors/destructors are needed to make forward declarations happy. LearningAgentsPolicy.h  
  LearningAgentsPolicy.h  

Destructors

Name Remarks Include Path Unreal Specifiers

virtual ~ULearningAgentsPolicy()

   LearningAgentsPolicy.h  

Variables

Protected

Name Type Remarks Include Path Unreal Specifiers
ActionDistributionVectors TLearningArray< 2, float > Action distribution vector buffers LearningAgentsPolicy.h  
ActionVectorEncodedIteration TLearningArray< 1, uint64, TInlineAllocator< 32 > > Number of times encoded action vector has been set for all agents LearningAgentsPolicy.h  
ActionVectorsEncoded TLearningArray< 2, float > Encoded action vector buffers LearningAgentsPolicy.h  
DecoderNetwork TObjectPtr< ULearningAgentsNeuralNetwork > The underlying decoder neural network. LearningAgentsPolicy.h
  • VisibleAnywhere
  • Transient
  • Category="LearningAgents"
DecoderObject TSharedPtr< UE::Learning::FNeuralNetworkFunction > Internal Decoder Function Object LearningAgentsPolicy.h  
EncoderNetwork TObjectPtr< ULearningAgentsNeuralNetwork > The underlying encoder neural network. LearningAgentsPolicy.h
  • VisibleAnywhere
  • Transient
  • Category="LearningAgents"
EncoderObject TSharedPtr< UE::Learning::FNeuralNetworkFunction > Internal Encoder Function Object LearningAgentsPolicy.h  
GlobalSeed uint32 The global random seed state used to initialize network weights and during sampling LearningAgentsPolicy.h  
Interactor TObjectPtr< ULearningAgentsInteractor > --Private Data -- LearningAgentsPolicy.h
  • VisibleAnywhere
  • Transient
  • Category="LearningAgents"
MemoryState TLearningArray< 2, float > The internal memory state of each agent. LearningAgentsPolicy.h  
MemoryStateIteration TLearningArray< 1, uint64, TInlineAllocator< 32 > > Number of times the memory state has been updated for all agents. LearningAgentsPolicy.h  
ObservationVectorEncodedIteration TLearningArray< 1, uint64, TInlineAllocator< 32 > > Number of times encoded observation vector has been set for all agents LearningAgentsPolicy.h  
ObservationVectorsEncoded TLearningArray< 2, float > Encoded observation vector buffers LearningAgentsPolicy.h  
PolicyNetwork TObjectPtr< ULearningAgentsNeuralNetwork > The underlying policy neural network. LearningAgentsPolicy.h
  • VisibleAnywhere
  • Transient
  • Category="LearningAgents"
PolicyObject TSharedPtr< UE::Learning::FNeuralNetworkPolicy > Internal Policy Function Object LearningAgentsPolicy.h  
PolicySettings FLearningAgentsPolicySettings   LearningAgentsPolicy.h  
PreEvaluationMemoryState TLearningArray< 2, float > The internal memory state of each agent before evaluation. LearningAgentsPolicy.h  
Seeds TLearningArray< 1, uint32 > The individual per-agent seeds used during sampling of actions. LearningAgentsPolicy.h  
ValidAgentIds TArray< int32 > Temp buffers used to record the set of agents that are valid for evaluation LearningAgentsPolicy.h  
ValidAgentSet UE::Learning::FIndexSet   LearningAgentsPolicy.h  

Functions

Public

Name Remarks Include Path Unreal Specifiers

TSharedRef< FJsonObject > AsJsonConfig()

   LearningAgentsPolicy.h  

void DecodeAndSampleActions ( const float ActionNoiseScale
)

 Decodes and samples action vectors using the Decoder network. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"
-- Blueprint public interface -- LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"

void EvaluatePolicy ()

 Calling this function will run the underlying neural network on the previously encoded observations to populate the encoded actions. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"

ULearningAgentsNeuralNetwork * GetDecoderNetworkAsset()

 Gets the current Encoder Network Asset being used LearningAgentsPolicy.h
  • BlueprintPure
  • Category="LearningAgents"
Get a reference to this policy's decoder function object. LearningAgentsPolicy.h  

ULearningAgentsNeuralNetwork * GetEncoderNetworkAsset()

 Gets the current Encoder Network Asset being used LearningAgentsPolicy.h
  • BlueprintPure
  • Category="LearningAgents"
-- Non-blueprint public interface -- LearningAgentsPolicy.h  

void GetMemoryState ( TArray< float >& OutMemoryState,
const int32 AgentId
) const

 Gets the current memory state for a given agent as represented by an abstract vector learned by the policy. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"
  • Meta=(AgentId=1)

TLearningArrayView< 2, const float > GetMemoryState ()

 Get a const view to this policy's memory state. LearningAgentsPolicy.h  

int32 GetMemoryStateSize()

 Gets the size of the memory state LearningAgentsPolicy.h
  • BlueprintPure
  • Category="LearningAgents"

TLearningArrayView< 2, const float > GetObservationVectorsEncoded()

 Get a const view to this policy's encoded observation vectors. LearningAgentsPolicy.h  

ULearningAgentsNeuralNetwork * GetPolicyNetworkAsset()

 Gets the current Policy Network Asset being used LearningAgentsPolicy.h
  • BlueprintPure
  • Category="LearningAgents"
Get a reference to this policy's policy function object. LearningAgentsPolicy.h  

TLearningArrayView< 2, const float > GetPreEvaluationMemoryState()

 Get a const view to this policy's pre-evaluation memory state. LearningAgentsPolicy.h  

bool HasEncodedObservationsForAgent ( const int32 AgentId
) const

 Returns true if this policy has encoded observations for the given agent id. Otherwise, false. LearningAgentsPolicy.h  

virtual void OnAgentsAdded_Implementation ( const TArray< int32 >& AgentIds
)

   LearningAgentsPolicy.h  

virtual void OnAgentsRemoved_Implementation ( const TArray< int32 >& AgentIds
)

   LearningAgentsPolicy.h  

virtual void OnAgentsReset_Implementation ( const TArray< int32 >& AgentIds
)

   LearningAgentsPolicy.h  

void RunInference ( const float ActionNoiseScale
)

 Calls GatherObservations, EncodeObservations, EvaluatePolicy, DecodeAndSampleActions, PerformActions LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"

void SetMemoryState ( const int32 AgentId,
const TArray< float >& InMemoryState
)

 Sets the current memory state for a given agent as represented by an abstract vector learned by the policy. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"
  • Meta=(AgentId=1)

void SetupPolicy ( ULearningAgentsManager*& InManager,
ULearningAgentsInteractor*& InInteractor,
ULearningAgentsNeuralNetwork* EncoderNeuralNetworkAsset,
ULearningAgentsNeuralNetwork* PolicyNeuralNetworkAsset,
ULearningAgentsNeuralNetwork* DecoderNeuralNetworkAsset,
const bool bReinitializeEncoderNetwork,
const bool bReinitializePolicyNetwork,
const bool bReinitializeDecoderNetwork,
const FLearningAgentsPolicySettings& InPolicySettings,
const int32 Seed
)

 Initializes this object to be used with the given agent interactor and policy settings. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"
  • Meta=(AutoCreateRefTerm="PolicySettings")

Static

Name Remarks Include Path Unreal Specifiers

static ULearningAgentsPolicy * MakePolicy ( ULearningAgentsManager*& InManager,
ULearningAgentsInteractor*& InInteractor,
TSubclassOf< ULearningAgentsPolicy > Class,
const FName Name,
ULearningAgentsNeuralNetwork* EncoderNeuralNetworkAsset,
ULearningAgentsNeuralNetwork* PolicyNeuralNetworkAsset,
ULearningAgentsNeuralNetwork* DecoderNeuralNetworkAsset,
const bool bReinitializeEncoderNetwork,
const bool bReinitializePolicyNetwork,
const bool bReinitializeDecoderNetwork,
const FLearningAgentsPolicySettings& PolicySettings,
const int32 Seed
)

 Constructs this object to be used with the given agent interactor and policy settings. LearningAgentsPolicy.h
  • BlueprintCallable
  • Category="LearningAgents"
  • Meta=(Class="/Script/LearningAgents.LearningAgentsPolicy", DeterminesOutputType="Class", AutoCreateRefTerm="PolicySettings")

Ask questions and help your peers Developer Forums

Write your own tutorials or read those from others Learning Library

On this page