opennn::Loss Class Reference

Unified loss container supporting MSE, cross-entropy, Minkowski, weighted, and regularized variants. More...

#include <loss.h>

Classes
struct	EvaluationResult
	Result of calculate_error; accuracy and active_tokens_count are populated only by classification losses. More...

Public Types
enum class	Error {   MeanSquaredError , NormalizedSquaredError , WeightedSquaredError , CrossEntropy ,   CrossEntropy3d , MinkowskiError }
	Error function selector used to dispatch the loss kernel. More...
enum class	Regularization { L1 , L2 , ElasticNet , NoRegularization }
	Parameter regularization method applied on top of the base loss. More...

Public Member Functions
	Loss (NeuralNetwork *=nullptr, Dataset *=nullptr)
	Constructs a Loss bound to an optional neural network and dataset.
virtual	~Loss ()=default
const NeuralNetwork *	get_neural_network () const
NeuralNetwork *	get_neural_network ()
const Dataset *	get_dataset () const
Dataset *	get_dataset ()
void	set (NeuralNetwork *=nullptr, Dataset *=nullptr)
	Resets the bound neural network and dataset pointers.
void	set_neural_network (NeuralNetwork *new_neural_network)
virtual void	set_dataset (Dataset *new_dataset)
void	set_regularization (const string &new_regularization_method)
void	set_regularization (Regularization new_regularization)
void	set_regularization_weight (const float new_regularization_weight)
void	set_normalization_coefficient ()
	Recomputes the normalization coefficient (used by NormalizedSquaredError) from the dataset.
EvaluationResult	calculate_error (const Batch &, const ForwardPropagation &) const
	Computes the loss for one batch using the cached forward-pass outputs.
void	set_error (const Error &)
	Selects the error function variant.
void	set_error (const string &)
	Selects the error function variant from its string name.
Error	get_error () const
void	back_propagate (const Batch &, ForwardPropagation &, BackPropagation &) const
	Performs the full backward pass: output deltas, layer gradients, and regularization gradient.
float	calculate_regularization (const VectorR &) const
	Returns the regularization penalty (L1, L2, or ElasticNet) for the given parameter vector.
void	from_JSON (const JsonDocument &)
	Restores loss configuration (error type, regularization, weights) from a JSON document.
void	to_JSON (JsonWriter &) const
	Serializes the loss configuration (error type, regularization, weights) to JSON.
void	regularization_from_JSON (const JsonDocument &)
	Restores the regularization sub-configuration from JSON.
void	regularization_to_JSON (JsonWriter &) const
	Serializes the regularization sub-configuration to JSON.
const string &	get_name () const
void	print () const
	Prints a human-readable description of the loss (no-op default).

Static Public Member Functions
static const EnumMap< Regularization > &	regularization_map ()
	Returns the static string<->enum map used to (de)serialize regularization types.
static const string &	regularization_to_string (Regularization regularization)
	Returns the canonical string name for a Regularization value.
static Regularization	string_to_regularization (const string &name)
	Parses a regularization name (accepts both "NoRegularization" and "None") back to the enum.
static float	calculate_h (const float)
	Returns the finite-difference step size h tuned for the given parameter value.

Protected Attributes
Error	error = Error::MeanSquaredError
float	normalization_coefficient = 1.0f
float	positives_weight = 1.0f
float	negatives_weight = 1.0f
float	minkowski_parameter = 1.5f
Regularization	regularization_method = Regularization::L2
float	regularization_weight = 0.001f
NeuralNetwork *	neural_network = nullptr
Dataset *	dataset = nullptr
string	name = "Loss"

Detailed Description

Unified loss container supporting MSE, cross-entropy, Minkowski, weighted, and regularized variants.

Member Enumeration Documentation

Error

enum class opennn::Loss::Error

strong

Error function selector used to dispatch the loss kernel.

Enumerator
MeanSquaredError
NormalizedSquaredError
WeightedSquaredError
CrossEntropy
CrossEntropy3d
MinkowskiError

Regularization

enum class opennn::Loss::Regularization

strong

Parameter regularization method applied on top of the base loss.

Enumerator
L1
L2
ElasticNet
NoRegularization

Constructor & Destructor Documentation

Loss()

opennn::Loss::Loss	(	NeuralNetwork *	= nullptr,
		Dataset *	= nullptr )

Constructs a Loss bound to an optional neural network and dataset.

Parameters

neural_network	Network whose outputs are scored (may be null and set later).
dataset	Dataset providing batches for error computation (may be null and set later).

~Loss()

virtual opennn::Loss::~Loss ( )

virtualdefault

Member Function Documentation

back_propagate()

void opennn::Loss::back_propagate	(	const Batch &	,
		ForwardPropagation &	,
		BackPropagation &	) const

Performs the full backward pass: output deltas, layer gradients, and regularization gradient.

Parameters

batch	The minibatch containing inputs and targets.
forward_propagation	Cached forward-pass state (may be updated for autograd-style layers).
back_propagation	Output structure receiving error and parameter gradients.

calculate_error()

EvaluationResult opennn::Loss::calculate_error	(	const Batch &	,
		const ForwardPropagation &	) const

Computes the loss for one batch using the cached forward-pass outputs.

Parameters

batch	The minibatch containing inputs and targets.
forward_propagation	Cached forward-pass state for this batch.

Returns

Loss value and, for classification losses, accuracy and active-token count.

calculate_h()

static float opennn::Loss::calculate_h ( const float )

static

Returns the finite-difference step size h tuned for the given parameter value.

calculate_regularization()

float opennn::Loss::calculate_regularization

(

const VectorR &

)

const

Returns the regularization penalty (L1, L2, or ElasticNet) for the given parameter vector.

from_JSON()

void opennn::Loss::from_JSON

(

const JsonDocument &

)

Restores loss configuration (error type, regularization, weights) from a JSON document.

get_dataset() [1/2]

Dataset * opennn::Loss::get_dataset ( )

inline

get_dataset() [2/2]

const Dataset * opennn::Loss::get_dataset ( ) const

inline

get_error()

Error opennn::Loss::get_error ( ) const

inline

get_name()

const string & opennn::Loss::get_name ( ) const

inline

get_neural_network() [1/2]

NeuralNetwork * opennn::Loss::get_neural_network ( )

inline

get_neural_network() [2/2]

const NeuralNetwork * opennn::Loss::get_neural_network ( ) const

inline

print()

void opennn::Loss::print ( ) const

inline

Prints a human-readable description of the loss (no-op default).

regularization_from_JSON()

void opennn::Loss::regularization_from_JSON

(

const JsonDocument &

)

Restores the regularization sub-configuration from JSON.

regularization_map()

static const EnumMap< Regularization > & opennn::Loss::regularization_map ( )

inlinestatic

Returns the static string<->enum map used to (de)serialize regularization types.

regularization_to_JSON()

void opennn::Loss::regularization_to_JSON

(

JsonWriter &

)

const

Serializes the regularization sub-configuration to JSON.

regularization_to_string()

static const string & opennn::Loss::regularization_to_string ( Regularization regularization )

inlinestatic

Returns the canonical string name for a Regularization value.

set()

void opennn::Loss::set	(	NeuralNetwork *	= nullptr,
		Dataset *	= nullptr )

Resets the bound neural network and dataset pointers.

set_dataset()

virtual void opennn::Loss::set_dataset ( Dataset * new_dataset )

inlinevirtual

set_error() [1/2]

void opennn::Loss::set_error

(

const Error &

)

Selects the error function variant.

set_error() [2/2]

void opennn::Loss::set_error

(

const string &

)

Selects the error function variant from its string name.

set_neural_network()

void opennn::Loss::set_neural_network ( NeuralNetwork * new_neural_network )

inline

set_normalization_coefficient()

void opennn::Loss::set_normalization_coefficient

(

)

Recomputes the normalization coefficient (used by NormalizedSquaredError) from the dataset.

set_regularization() [1/2]

void opennn::Loss::set_regularization ( const string & new_regularization_method )

inline

set_regularization() [2/2]

void opennn::Loss::set_regularization ( Regularization new_regularization )

inline

set_regularization_weight()

void opennn::Loss::set_regularization_weight ( const float new_regularization_weight )

inline

string_to_regularization()

static Regularization opennn::Loss::string_to_regularization ( const string & name )

inlinestatic

Parses a regularization name (accepts both "NoRegularization" and "None") back to the enum.

to_JSON()

void opennn::Loss::to_JSON

(

JsonWriter &

)

const

Serializes the loss configuration (error type, regularization, weights) to JSON.

Member Data Documentation

dataset

Dataset* opennn::Loss::dataset = nullptr

protected

error

Error opennn::Loss::error = Error::MeanSquaredError

protected

minkowski_parameter

float opennn::Loss::minkowski_parameter = 1.5f

protected

name

string opennn::Loss::name = "Loss"

protected

negatives_weight

float opennn::Loss::negatives_weight = 1.0f

protected

neural_network

NeuralNetwork* opennn::Loss::neural_network = nullptr

protected

normalization_coefficient

float opennn::Loss::normalization_coefficient = 1.0f

protected

positives_weight

float opennn::Loss::positives_weight = 1.0f

protected

regularization_method

Regularization opennn::Loss::regularization_method = Regularization::L2

protected

regularization_weight

float opennn::Loss::regularization_weight = 0.001f

protected