UInferencePolicy

class UInferencePolicy : public UAbstractPolicy

一个使用训练好的 NNE 模型进行决策的策略。

依赖项

此类型依赖于

• FDictPoint
• FDictSpace
• FInferencePolicyBuffer
• FInteractionDefinition
• FPolicyDecision
• IRuntimeInterface
• UAbstractPolicy

继承自: public UAbstractPolicy

公共接口

构造函数

UInferencePolicy

UInferencePolicy()

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 200, column 1)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 13-15)

公共函数

GetRuntimeNames

TArray<FString> GetRuntimeNames() const const

获取所有可用运行时名称的函数。

返回: 可用运行时名称的数组

返回: TArray<FString>

属性: const

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 227, column 8)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 17-23)

GetRuntime

IRuntimeInterface * GetRuntime(const FString &SelectedRuntimeName) const const

获取具有给定名称的运行时的单例实例的函数。

返回: 指向具有给定名称的运行时的指针

参数

• SelectedRuntimeName (const FString &) – [in] 要获取的运行时的名称

返回: IRuntimeInterface *

属性: const

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 234, column 19)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 25-40)

RequestDecision

virtual TFuture<FPolicyDecision *> RequestDecision(const FDictPoint &Observations) override

请求策略做出行动决策。

可能需要一些时间才能发生

返回: 一个 Future，最终将包含策略的下一个决策

参数

• Observations (const FDictPoint &) – [输入] 代理的当前状态，用于告知策略的选择的动作

返回: TFuture<FPolicyDecision *>

属性: virtual

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 240, column 17)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 43-94)

初始化

virtual void Init(const FInteractionDefinition &PolicyDefinition)

根据交互定义初始化策略对象实例。

参数

• PolicyDefinition (const FInteractionDefinition &) – [输入] 定义策略的 I/O 形状和其他参数的对象

属性: virtual

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 242, column 6)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 250-303)

SetupBuffersAndBindings

bool SetupBuffersAndBindings(const FInteractionDefinition &PolicyDefinition, TSharedPtr<IModelInstanceRunSync> ModelInstance)

参数

名称	类型	默认
PolicyDefinition	const FInteractionDefinition &	“
ModelInstance	TSharedPtr<IModelInstanceRunSync>	“

返回: bool

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 244, column 6)

实现: Schola/Source/Schola/Private/Policies/InferencePolicy.cpp (lines 135-247)

公共成员

int Step

int Step = = 0

对策略的推理调用次数。

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 204, column 5)

FDictSpace ActionSpaceDefn

FDictSpace ActionSpaceDefn

代理的动作空间。

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 208, column 12)

FDictSpace ObservationSpaceDefn

FDictSpace ObservationSpaceDefn

代理的观察空间。

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 212, column 12)

TObjectPtr<UNNEModelData> ModelData

TObjectPtr<UNNEModelData> ModelData

NNE 模型的模型数据。

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 216, column 12)

FString RuntimeName

FString RuntimeName

用于推理的运行时名称。

如果没有可用的运行时，则启用 NNE 插件（例如 NNERuntimeORT）。

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 220, column 9)

bool bNetworkLoaded

bool bNetworkLoaded = = false

跟踪网络是否正确加载的变量？

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 238, column 6)

TArray<FInferencePolicyBuffer> ActionBuffer

TArray<FInferencePolicyBuffer> ActionBuffer

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 247, column 8)

TArray<FInferencePolicyBuffer> ObservationBuffer

TArray<FInferencePolicyBuffer> ObservationBuffer

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 250, column 8)

TArray<float> StateBuffer

TArray<float> StateBuffer

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 253, column 8)

int StateSeqLen

int StateSeqLen

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 256, column 5)

int StateDimSize

int StateDimSize

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 259, column 5)

私有接口

私有成员

TSharedPtr<IModelInstanceRunSync> ModelInstance

TSharedPtr<IModelInstanceRunSync> ModelInstance

已实例化的模型。

属性: private

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 263, column 12)

TArray<UE::NNE::FTensorBindingCPU> InputBindings

TArray<UE::NNE::FTensorBindingCPU> InputBindings

属性: private

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 265, column 8)

TArray<UE::NNE::FTensorBindingCPU> OutputBindings

TArray<UE::NNE::FTensorBindingCPU> OutputBindings

属性: private

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 267, column 8)

来源: Schola/Source/Schola/Public/Policies/InferencePolicy.h (line 195, column 1)