Deep::DiscriminativePCNetwork Class Reference

An abstracted class for an array of DiscriminativePCLayer. More...

#include <DiscriminativePCNetwork.h>

Public Member Functions
	DiscriminativePCNetwork ()
	Default constructor.
	DiscriminativePCNetwork (int batchSize)
	Batched constructor.
	~DiscriminativePCNetwork ()=default
	Default constructor; deletes each layer.
	DiscriminativePCNetwork (DiscriminativePCNetwork &&other) noexcept=default
	Move constructor.
DiscriminativePCNetwork &	operator= (DiscriminativePCNetwork &&other) noexcept=default
	Move assignment operator.
void	AddLayer (int size, int nextSize, float lr, float ir, float pr, float lmbda, void(*act)(float *, size_t), void(*dAct)(float *, size_t, bool))
	Adds a layer to the network.
void	AddLayer (int size, int nextSize, float lr, float ir, float pr, float lmbda, Deep::ActivationType aType, Deep::ActivationType dType)
	Adds a layer to the network, using a named ActivationType instead of raw function pointers.
void	RandomizeWeights (std::mt19937 &rng)
	Randomizes the weights of each layer.
void	Clamp (const std::vector< float > &input)
	Clamps the input to the first layer, necessary for prediction.
float	CalculateState ()
	Calculates the state of each layer.
void	UpdateState ()
	Updates each layer's state.
void	UpdateWeights ()
	Updates each layer's weights.
void	UpdatePrecision ()
	Updates each layer's precision weighting.
void	ProjectForward () noexcept
	Seeds every hidden layer's belief from a genuine forward pass through current weights, rather than zero-init. Assumes the input layer is already clamped.
float	TrainStepWithProjection (const std::vector< float > &x, const std::vector< float > &y, int inferenceSteps)
	Same as TrainStep(), but with ProjectForward() called once after clamping the input, before the settling loop.
std::vector< float >	PredictWithProjection (const std::vector< float > &x, int inferenceSteps)
	Same as Predict(), but with ProjectForward() called once after clamping the input, before the settling loop.
void	ResetState () noexcept
	Resets each layer's state without touching learned weights.
const auto &	GetLayers () const noexcept
	Returns every layer in the network, in the order they were added.
int	GetBatchSize () const noexcept
	Returns the batch size for the network's layers.
DiscriminativePCLayer *	GetTerminalLayer () const
	Returns the network's terminal (final) layer.
void	SetLearningRate (float lr)
	Sets the learning rate used for weight updates, on every layer.
void	SetInferenceRate (float ir)
	Sets the inference rate used for state updates, on every layer.
void	SetPrecisionRate (float pr)
	Sets the learning rate used for precision updates, on every layer.
void	SetLambda (float l)
	Sets the weight-decay (L2 regularization) coefficient, on every layer.
void	SetOptimizer (OptimizerType opt) noexcept
	Sets the optimizer used for weight updates, on every layer.
float	TrainStep (const std::vector< float > &x, const std::vector< float > &y, int inferenceSteps)
	Runs a complete training step (clamp, settle, update, unclamp).
std::vector< float >	Predict (const std::vector< float > &x, int inferenceSteps)
	Runs a forward prediction pass (clamp, settle, read).
bool	Save (const std::string &filename) const noexcept
	Saves the network's learned parameters to disk.
bool	Load (const std::string &filename) noexcept
	Loads the network's learned parameters from disk.
void	Compile ()
	Loads all layers into one contiguous block of memory.

Friends
class	PCNDiagnostics

Detailed Description

An abstracted class for an array of DiscriminativePCLayer.

See also

https://arxiv.org/pdf/2506.06332

Constructor & Destructor Documentation

DiscriminativePCNetwork() [1/3]

Deep::DiscriminativePCNetwork::DiscriminativePCNetwork ( )

inline

Default constructor.

Initializes the network with auto-batch size detection.

DiscriminativePCNetwork() [2/3]

Deep::DiscriminativePCNetwork::DiscriminativePCNetwork ( int batchSize )

inline

Batched constructor.

Parameters

batchSize

Batch size

Initializes the network with a predetermined batch size.

~DiscriminativePCNetwork()

Deep::DiscriminativePCNetwork::~DiscriminativePCNetwork ( )

default

Default constructor; deletes each layer.

DiscriminativePCNetwork() [3/3]

Deep::DiscriminativePCNetwork::DiscriminativePCNetwork ( DiscriminativePCNetwork && other )

defaultnoexcept

Move constructor.

Member Function Documentation

AddLayer() [1/2]

void Deep::DiscriminativePCNetwork::AddLayer	(	int	size,
		int	nextSize,
		float	lr,
		float	ir,
		float	pr,
		float	lmbda,
		Deep::ActivationType	aType,
		Deep::ActivationType	dType )

Adds a layer to the network, using a named ActivationType instead of raw function pointers.

Parameters

size	input size
nextSize	output size
lr	learning rate for beliefs
ir	learning rate for weights
pr	learning rate for precision
lmbda	weight decay (L2 regularization) coefficient
aType	activation type
dType	activation derivative type

AddLayer() [2/2]

void Deep::DiscriminativePCNetwork::AddLayer	(	int	size,
		int	nextSize,
		float	lr,
		float	ir,
		float	pr,
		float	lmbda,
		void(*	act )(float *, size_t),
		void(*	dAct )(float *, size_t, bool) )

Adds a layer to the network.

Parameters

size	input size
nextSize	output size
lr	learning rate for beliefs
ir	learning rate for weights
pr	learning rate for precision
lmbda	weight decay (L2 regularization) coefficient
act	activation function
dAct	derivative of previous activation function

CalculateState()

float Deep::DiscriminativePCNetwork::CalculateState

(

)

Calculates the state of each layer.

Returns

Returns total energy

Clamp()

void Deep::DiscriminativePCNetwork::Clamp

(

const std::vector< float > &

input

)

Clamps the input to the first layer, necessary for prediction.

Parameters

input

reference to input vector

Compile()

void Deep::DiscriminativePCNetwork::Compile

(

)

Loads all layers into one contiguous block of memory.

GetBatchSize()

int Deep::DiscriminativePCNetwork::GetBatchSize ( ) const

inlinenoexcept

Returns the batch size for the network's layers.

Returns

size_t batchSize

GetLayers()

const auto & Deep::DiscriminativePCNetwork::GetLayers ( ) const

inlinenoexcept

Returns every layer in the network, in the order they were added.

Returns

A const reference to the internal layer list.

GetTerminalLayer()

DiscriminativePCLayer * Deep::DiscriminativePCNetwork::GetTerminalLayer ( ) const

inline

Returns the network's terminal (final) layer.

Returns

Pointer to the last layer added via AddLayer(), or nullptr if no layers have been added.

Load()

bool Deep::DiscriminativePCNetwork::Load ( const std::string & filename )

noexcept

Loads the network's learned parameters from disk.

Parameters

filename

The source file path.

Returns

True on success, false otherwise.

operator=()

DiscriminativePCNetwork & Deep::DiscriminativePCNetwork::operator= ( DiscriminativePCNetwork && other )

defaultnoexcept

Move assignment operator.

Predict()

std::vector< float > Deep::DiscriminativePCNetwork::Predict	(	const std::vector< float > &	x,
		int	inferenceSteps )

Runs a forward prediction pass (clamp, settle, read).

Parameters

x	The batched input data
inferenceSteps	The number of relaxation iterations

Returns

A vector containing the batched predictions

PredictWithProjection()

std::vector< float > Deep::DiscriminativePCNetwork::PredictWithProjection	(	const std::vector< float > &	x,
		int	inferenceSteps )

Same as Predict(), but with ProjectForward() called once after clamping the input, before the settling loop.

Parameters

x	The batched input data
inferenceSteps	The number of relaxation iterations

Returns

A vector containing the batched predictions

ProjectForward()

void Deep::DiscriminativePCNetwork::ProjectForward ( )

noexcept

Seeds every hidden layer's belief from a genuine forward pass through current weights, rather than zero-init. Assumes the input layer is already clamped.

RandomizeWeights()

void Deep::DiscriminativePCNetwork::RandomizeWeights

(

std::mt19937 &

rng

)

Randomizes the weights of each layer.

Parameters

rng	The classic Mersenne Twister

ResetState()

void Deep::DiscriminativePCNetwork::ResetState ( )

noexcept

Resets each layer's state without touching learned weights.

Save()

bool Deep::DiscriminativePCNetwork::Save ( const std::string & filename ) const

noexcept

Saves the network's learned parameters to disk.

Parameters

filename

The destination file path.

Returns

True on success, false otherwise.

SetInferenceRate()

void Deep::DiscriminativePCNetwork::SetInferenceRate ( float ir )

inline

Sets the inference rate used for state updates, on every layer.

Parameters

ir	The new inference rate.

SetLambda()

void Deep::DiscriminativePCNetwork::SetLambda ( float l )

inline

Sets the weight-decay (L2 regularization) coefficient, on every layer.

Parameters

l	The new lambda value.

SetLearningRate()

void Deep::DiscriminativePCNetwork::SetLearningRate ( float lr )

inline

Sets the learning rate used for weight updates, on every layer.

Parameters

lr	The new learning rate.

SetOptimizer()

void Deep::DiscriminativePCNetwork::SetOptimizer ( OptimizerType opt )

inlinenoexcept

Sets the optimizer used for weight updates, on every layer.

Parameters

opt	The optimizer type to apply.

SetPrecisionRate()

void Deep::DiscriminativePCNetwork::SetPrecisionRate ( float pr )

inline

Sets the learning rate used for precision updates, on every layer.

Parameters

pr	The new precision rate.

TrainStep()

float Deep::DiscriminativePCNetwork::TrainStep	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		int	inferenceSteps )

Runs a complete training step (clamp, settle, update, unclamp).

Parameters

x	The batched input data
y	The batched target data
inferenceSteps	The number of relaxation iterations

Returns

The final energy state of the network before weight updates

TrainStepWithProjection()

float Deep::DiscriminativePCNetwork::TrainStepWithProjection	(	const std::vector< float > &	x,
		const std::vector< float > &	y,
		int	inferenceSteps )

Same as TrainStep(), but with ProjectForward() called once after clamping the input, before the settling loop.

Parameters

x	The batched input data
y	The batched target data
inferenceSteps	The number of relaxation iterations

Returns

The final energy state of the network before weight updates

UpdatePrecision()

void Deep::DiscriminativePCNetwork::UpdatePrecision

(

)

Updates each layer's precision weighting.

UpdateState()

void Deep::DiscriminativePCNetwork::UpdateState

(

)

Updates each layer's state.

UpdateWeights()

void Deep::DiscriminativePCNetwork::UpdateWeights

(

)

Updates each layer's weights.

Friends And Related Symbol Documentation

PCNDiagnostics

friend class PCNDiagnostics

friend

---

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

• include/deepity/networks/DiscriminativePCNetwork.h
• src/networks/DiscriminativePCNetwork.cpp