This is a basic multilayer perceptron neural network capable of basic training and feed forward operations. More...

#include <MultilayerPerceptron.h>

Inherits MachineLearning::INeuralNetwork.

Public Member Functions
	AZ_RTTI (MultilayerPerceptron, "{E12EF761-41A5-48C3-BF55-7179B280D45F}", INeuralNetwork)

	MultilayerPerceptron (const MultilayerPerceptron &)

	MultilayerPerceptron (AZStd::size_t activationCount)

MultilayerPerceptron &	operator= (const MultilayerPerceptron &)

MultilayerPerceptron &	operator= (const ModelAsset &)

void	AddLayer (AZStd::size_t layerDimensionality, ActivationFunctions activationFunction=ActivationFunctions::ReLU)
	Adds a new layer to the model.

Layer *	GetLayer (AZStd::size_t layerIndex)
	Retrieves a specific layer from the model, this is not thread safe and should only be used during unit testing to validate model parameters.


AZStd::string	GetName () const override

AZStd::string	GetAssetFile (AssetTypes assetType) const override

AZStd::size_t	GetInputDimensionality () const override
	Returns the number of input neurons the model supports.

AZStd::size_t	GetOutputDimensionality () const override
	Returns the number of output neurons the model supports.

AZStd::size_t	GetLayerCount () const override
	Returns the total number of layers in the network.

AZ::MatrixMxN	GetLayerWeights (AZStd::size_t layerIndex) const override

AZ::VectorN	GetLayerBiases (AZStd::size_t layerIndex) const override

AZStd::size_t	GetParameterCount () const override
	Returns the total number of parameters in the neural network.

IInferenceContextPtr	CreateInferenceContext () override

ITrainingContextPtr	CreateTrainingContext () override

const AZ::VectorN *	Forward (IInferenceContextPtr context, const AZ::VectorN &activations) override

void	Reverse (ITrainingContextPtr context, LossFunctions lossFunction, const AZ::VectorN &activations, const AZ::VectorN &expected) override

void	GradientDescent (ITrainingContextPtr context, float learningRate) override

bool	LoadModel () override
	Loads the current model parameters from the associated asset file.

bool	SaveModel () override
	Saves the current model parameters to the associated asset file.

Public Member Functions inherited from MachineLearning::INeuralNetwork
	AZ_RTTI (INeuralNetwork, "{64E5B5B1-4A7D-489D-9A29-D9510BB7E17A}")

	INeuralNetwork (INeuralNetwork &&)=default

	INeuralNetwork (const INeuralNetwork &)=default

INeuralNetwork &	operator= (INeuralNetwork &&)=default

INeuralNetwork &	operator= (const INeuralNetwork &)=default

virtual AZStd::string	GetAssetFile ([[maybe_unused]] AssetTypes assetType) const
	Returns the file where model parameters are stored.

virtual AZ::MatrixMxN	GetLayerWeights ([[maybe_unused]] AZStd::size_t layerIndex) const
	Returns the weight matrix for the requested layer.

virtual AZ::VectorN	GetLayerBiases ([[maybe_unused]] AZStd::size_t layerIndex) const
	Returns the weight matrix for the requested layer.

virtual const AZ::VectorN *	Forward ([[maybe_unused]] IInferenceContextPtr context, [[maybe_unused]] const AZ::VectorN &activations)
	Performs a basic feed-forward operation to compute the output from a set of activation values.

virtual void	Reverse ([[maybe_unused]] ITrainingContextPtr context, [[maybe_unused]] LossFunctions lossFunction, [[maybe_unused]] const AZ::VectorN &activations, [[maybe_unused]] const AZ::VectorN &expected)
	Accumulates the loss gradients given a loss function, an activation vector and a corresponding label vector.

virtual void	GradientDescent ([[maybe_unused]] ITrainingContextPtr context, [[maybe_unused]] float learningRate)
	Performs a gradient descent step and resets all gradient accumulators to zero.

void	add_ref ()

void	release ()

Static Public Member Functions
static void	Reflect (AZ::ReflectContext *context)

Friends
class	MultilayerPerceptronEditorComponent

Detailed Description

This is a basic multilayer perceptron neural network capable of basic training and feed forward operations.

Member Function Documentation

◆ GetName()

AZStd::string MachineLearning::MultilayerPerceptron::GetName ( ) const

overridevirtual

INeuralNetwork interface

Reimplemented from MachineLearning::INeuralNetwork.

◆ Reflect()

static void MachineLearning::MultilayerPerceptron::Reflect ( AZ::ReflectContext * context )

static

AzCore Reflection.

Parameters

context reflection context

The documentation for this class was generated from the following file:

Gems/MachineLearning/Code/Source/Models/MultilayerPerceptron.h

Public Member Functions

Static Public Member Functions

Friends

Detailed Description

Member Function Documentation

◆ GetName()

◆ Reflect()