Loss function interface. More...

#include <shark/ObjectiveFunctions/Loss/AbstractLoss.h>

Inheritance diagram for shark::AbstractLoss< LabelT, OutputT >:

Public Types
typedef OutputT	OutputType

typedef LabelT	LabelType

typedef RealMatrix	MatrixType

typedef Batch< OutputType >::type	BatchOutputType

typedef Batch< LabelType >::type	BatchLabelType

typedef ConstProxyReference< LabelTypeconst >::type	ConstLabelReference
	Const references to LabelType.

typedef ConstProxyReference< OutputTypeconst >::type	ConstOutputReference
	Const references to OutputType.

Public Types inherited from shark::AbstractCost< LabelT, OutputT >
enum	Feature { HAS_FIRST_DERIVATIVE = 1 , HAS_SECOND_DERIVATIVE = 2 , IS_LOSS_FUNCTION = 4 }
	list of features a cost function can have More...

typedef OutputT	OutputType

typedef LabelT	LabelType

typedef Batch< OutputType >::type	BatchOutputType

typedef Batch< LabelType >::type	BatchLabelType

typedef TypedFlags< Feature >	Features

typedef TypedFeatureNotAvailableException< Feature >	FeatureNotAvailableException

Public Member Functions
	AbstractLoss ()

virtual double	eval (BatchLabelType const &target, BatchOutputType const &prediction) const =0
	evaluate the loss for a batch of targets and a prediction

virtual double	eval (ConstLabelReference target, ConstOutputReference prediction) const
	evaluate the loss for a target and a prediction

virtual double	evalDerivative (ConstLabelReference target, ConstOutputReference prediction, OutputType &gradient) const
	evaluate the loss and its derivative for a target and a prediction

virtual double	evalDerivative (ConstLabelReference target, ConstOutputReference prediction, OutputType &gradient, MatrixType &hessian) const
	evaluate the loss and its first and second derivative for a target and a prediction

virtual double	evalDerivative (BatchLabelType const &target, BatchOutputType const &prediction, BatchOutputType &gradient) const
	evaluate the loss and the derivative w.r.t. the prediction

double	eval (Data< LabelType > const &targets, Data< OutputType > const &predictions) const

double	operator() (LabelType const &target, OutputType const &prediction) const
	evaluate the loss for a target and a prediction

double	operator() (BatchLabelType const &target, BatchOutputType const &prediction) const

Public Member Functions inherited from shark::AbstractCost< LabelT, OutputT >
virtual	~AbstractCost ()

const Features &	features () const

virtual void	updateFeatures ()

bool	hasFirstDerivative () const
	returns true when the first parameter derivative is implemented

bool	isLossFunction () const
	returns true when the cost function is in fact a loss function

double	operator() (Data< LabelType > const &targets, Data< OutputType > const &predictions) const

Public Member Functions inherited from shark::INameable
virtual	~INameable ()

virtual std::string	name () const
	returns the name of the object

Additional Inherited Members
Protected Attributes inherited from shark::AbstractCost< LabelT, OutputT >
Features	m_features

Detailed Description

template<class LabelT, class OutputT = LabelT>
class shark::AbstractLoss< LabelT, OutputT >

Loss function interface.

: In statistics and machine learning, a loss function encodes the severity of getting a label wrong. This is am important special case of a cost function (see AbstractCost), where the cost is computed as the average loss over a set, also known as (empirical) risk.

: It is generally agreed that loss values are non-negative, and that the loss of correct prediction is zero. This rule is not formally checked, but instead left to the various sub-classes.

Definition at line 62 of file AbstractLoss.h.

Member Typedef Documentation

◆ BatchLabelType

template<class LabelT , class OutputT = LabelT>

typedef Batch<LabelType>::type shark::AbstractLoss< LabelT, OutputT >::BatchLabelType

Definition at line 71 of file AbstractLoss.h.

◆ BatchOutputType

template<class LabelT , class OutputT = LabelT>

typedef Batch<OutputType>::type shark::AbstractLoss< LabelT, OutputT >::BatchOutputType

Definition at line 70 of file AbstractLoss.h.

◆ ConstLabelReference

template<class LabelT , class OutputT = LabelT>

typedef ConstProxyReference<LabelTypeconst>::type shark::AbstractLoss< LabelT, OutputT >::ConstLabelReference

Const references to LabelType.

Definition at line 74 of file AbstractLoss.h.

◆ ConstOutputReference

template<class LabelT , class OutputT = LabelT>

typedef ConstProxyReference<OutputTypeconst>::type shark::AbstractLoss< LabelT, OutputT >::ConstOutputReference

Const references to OutputType.

Definition at line 76 of file AbstractLoss.h.

◆ LabelType

template<class LabelT , class OutputT = LabelT>

typedef LabelT shark::AbstractLoss< LabelT, OutputT >::LabelType

Definition at line 67 of file AbstractLoss.h.

◆ MatrixType

template<class LabelT , class OutputT = LabelT>

typedef RealMatrix shark::AbstractLoss< LabelT, OutputT >::MatrixType

Definition at line 68 of file AbstractLoss.h.

◆ OutputType

template<class LabelT , class OutputT = LabelT>

typedef OutputT shark::AbstractLoss< LabelT, OutputT >::OutputType

Definition at line 66 of file AbstractLoss.h.

Constructor & Destructor Documentation

◆ AbstractLoss()

template<class LabelT , class OutputT = LabelT>

shark::AbstractLoss< LabelT, OutputT >::AbstractLoss ( )

inline

Definition at line 78 of file AbstractLoss.h.

References shark::AbstractCost< LabelT, OutputT >::IS_LOSS_FUNCTION, and shark::AbstractCost< LabelT, OutputT >::m_features.

Member Function Documentation

◆ eval() [1/3]

template<class LabelT , class OutputT = LabelT>

virtual double shark::AbstractLoss< LabelT, OutputT >::eval	(	BatchLabelType const &	target,
		BatchOutputType const &	prediction
	)		const

pure virtual

evaluate the loss for a batch of targets and a prediction

Parameters

target	target values
prediction	predictions, typically made by a model

Implemented in shark::AbsoluteLoss< VectorType >, shark::EpsilonHingeLoss, shark::HingeLoss, shark::HuberLoss, shark::SquaredEpsilonHingeLoss, shark::SquaredHingeLoss, shark::SquaredLoss< OutputType, LabelType >, shark::SquaredLoss< OutputType, unsigned int >, shark::SquaredLoss< Sequence, Sequence >, shark::ZeroOneLoss< LabelType, OutputType >, shark::ZeroOneLoss< unsigned int, blas::vector< Float > >, shark::DiscreteLoss, shark::AbsoluteLoss< VectorType >, shark::CrossEntropy< blas::vector< T, Device >, blas::vector< T, Device > >, shark::SquaredLoss< OutputType, LabelType >, and shark::ZeroOneLoss< LabelType, OutputType >.

Referenced by shark::LooError< ModelTypeT, LabelType >::eval(), shark::AbstractLoss< LabelT, OutputT >::eval(), shark::AbstractLoss< LabelT, OutputT >::eval(), shark::AbstractLoss< LabelT, OutputT >::operator()(), and shark::AbstractLoss< LabelT, OutputT >::operator()().

◆ eval() [2/3]

template<class LabelT , class OutputT = LabelT>

virtual double shark::AbstractLoss< LabelT, OutputT >::eval	(	ConstLabelReference	target,
		ConstOutputReference	prediction
	)		const

inlinevirtual

evaluate the loss for a target and a prediction

Parameters

target	target value
prediction	prediction, typically made by a model

Reimplemented in shark::AbsoluteLoss< VectorType >, shark::CrossEntropy< unsigned int, OutputType >, shark::CrossEntropy< blas::vector< T, Device >, blas::vector< T, Device > >, shark::SquaredLoss< OutputType, LabelType >, and shark::ZeroOneLoss< LabelType, OutputType >.

Definition at line 92 of file AbstractLoss.h.

References shark::AbstractLoss< LabelT, OutputT >::eval(), and shark::getBatchElement().

◆ eval() [3/3]

template<class LabelT , class OutputT = LabelT>

double shark::AbstractLoss< LabelT, OutputT >::eval	(	Data< LabelType > const &	targets,
		Data< OutputType > const &	predictions
	)		const

inlinevirtual

from AbstractCost

Parameters

targets	target values
predictions	predictions, typically made by a model

Implements shark::AbstractCost< LabelT, OutputT >.

Reimplemented in shark::SquaredLoss< OutputType, LabelType >, and shark::ZeroOneLoss< LabelType, OutputType >.

Definition at line 149 of file AbstractLoss.h.

References shark::Data< Type >::batch(), shark::AbstractLoss< LabelT, OutputT >::eval(), shark::Data< Type >::numberOfBatches(), shark::Data< Type >::numberOfElements(), SHARK_CRITICAL_REGION, SHARK_PARALLEL_FOR, and SIZE_CHECK.

◆ evalDerivative() [1/3]

template<class LabelT , class OutputT = LabelT>

virtual double shark::AbstractLoss< LabelT, OutputT >::evalDerivative	(	BatchLabelType const &	target,
		BatchOutputType const &	prediction,
		BatchOutputType &	gradient
	)		const

inlinevirtual

evaluate the loss and the derivative w.r.t. the prediction

: The default implementations throws an exception. If you overwrite this method, don't forget to set the flag HAS_FIRST_DERIVATIVE.

Parameters

target	target value
prediction	prediction, typically made by a model
gradient	the gradient of the loss function with respect to the prediction

Reimplemented in shark::SquaredLoss< OutputType, LabelType >, shark::EpsilonHingeLoss, shark::HingeLoss, shark::HuberLoss, shark::SquaredEpsilonHingeLoss, shark::SquaredHingeLoss, shark::SquaredLoss< OutputType, unsigned int >, and shark::SquaredLoss< Sequence, Sequence >.

Definition at line 139 of file AbstractLoss.h.

References shark::AbstractCost< LabelT, OutputT >::HAS_FIRST_DERIVATIVE, and SHARK_FEATURE_EXCEPTION_DERIVED.

◆ evalDerivative() [2/3]

template<class LabelT , class OutputT = LabelT>

virtual double shark::AbstractLoss< LabelT, OutputT >::evalDerivative	(	ConstLabelReference	target,
		ConstOutputReference	prediction,
		OutputType &	gradient
	)		const

inlinevirtual

evaluate the loss and its derivative for a target and a prediction

Parameters

target	target value
prediction	prediction, typically made by a model
gradient	the gradient of the loss function with respect to the prediction

Reimplemented in shark::CrossEntropy< unsigned int, OutputType >.

Definition at line 105 of file AbstractLoss.h.

References shark::AbstractLoss< LabelT, OutputT >::evalDerivative(), and shark::getBatchElement().

Referenced by shark::AbstractLoss< LabelT, OutputT >::evalDerivative(), shark::VariationalAutoencoderError< SearchPointType >::evalDerivative(), and shark::KernelSGDTrainer< InputType, CacheType >::train().

◆ evalDerivative() [3/3]

template<class LabelT , class OutputT = LabelT>

virtual double shark::AbstractLoss< LabelT, OutputT >::evalDerivative	(	ConstLabelReference	target,
		ConstOutputReference	prediction,
		OutputType &	gradient,
		MatrixType &	hessian
	)		const

inlinevirtual

evaluate the loss and its first and second derivative for a target and a prediction

Parameters

target	target value
prediction	prediction, typically made by a model
gradient	the gradient of the loss function with respect to the prediction
hessian	the hessian of the loss function with respect to the prediction

Definition at line 122 of file AbstractLoss.h.

References shark::AbstractCost< LabelT, OutputT >::HAS_SECOND_DERIVATIVE, and SHARK_FEATURE_EXCEPTION_DERIVED.

◆ operator()() [1/2]

template<class LabelT , class OutputT = LabelT>

double shark::AbstractLoss< LabelT, OutputT >::operator()	(	BatchLabelType const &	target,
		BatchOutputType const &	prediction
	)		const

inline

Definition at line 173 of file AbstractLoss.h.

References shark::AbstractLoss< LabelT, OutputT >::eval().

◆ operator()() [2/2]

template<class LabelT , class OutputT = LabelT>

double shark::AbstractLoss< LabelT, OutputT >::operator()	(	LabelType const &	target,
		OutputType const &	prediction
	)		const

inline

evaluate the loss for a target and a prediction

: convenience operator

Parameters

target	target value
prediction	prediction, typically made by a model

Definition at line 170 of file AbstractLoss.h.

References shark::AbstractLoss< LabelT, OutputT >::eval().

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

include/shark/ObjectiveFunctions/Loss/AbstractLoss.h

Public Types

Public Member Functions

Additional Inherited Members

Detailed Description

Member Typedef Documentation

◆ BatchLabelType

◆ BatchOutputType

◆ ConstLabelReference

◆ ConstOutputReference

◆ LabelType

◆ MatrixType

◆ OutputType

Constructor & Destructor Documentation

◆ AbstractLoss()

Member Function Documentation

◆ eval() [1/3]

◆ eval() [2/3]

◆ eval() [3/3]

◆ evalDerivative() [1/3]

◆ evalDerivative() [2/3]

◆ evalDerivative() [3/3]

◆ operator()() [1/2]

◆ operator()() [2/2]