Implements a layer of radial basis functions in a neural network. More...

#include <shark/Models/RBFLayer.h>

Inheritance diagram for shark::RBFLayer:

Public Member Functions
SHARK_EXPORT_SYMBOL	RBFLayer ()
	Creates an empty Radial Basis Function layer.

SHARK_EXPORT_SYMBOL	RBFLayer (std::size_t numInput, std::size_t numOutput)
	Creates a layer of a Radial Basis Function Network.

std::string	name () const
	From INameable: return the class name.

SHARK_EXPORT_SYMBOL RealVector	parameterVector () const
	Returns the current parameter vector. The amount and order of weights depend on the training parameters.

SHARK_EXPORT_SYMBOL void	setParameterVector (RealVector const &newParameters)
	Sets the new internal parameters.

SHARK_EXPORT_SYMBOL std::size_t	numberOfParameters () const
	Returns the number of parameters which are currently enabled for training.

Shape	inputShape () const
	Returns the number of input neurons.

Shape	outputShape () const
	Returns the number of output neurons.

boost::shared_ptr< State >	createState () const
	Creates an internal state of the model.

SHARK_EXPORT_SYMBOL void	setStructure (std::size_t numInput, std::size_t numOutput)
	Configures a Radial Basis Function Network.

SHARK_EXPORT_SYMBOL void	eval (BatchInputType const &patterns, BatchOutputType &outputs, State &state) const
	Standard interface for evaluating the response of the model to a batch of patterns.

SHARK_EXPORT_SYMBOL void	weightedParameterDerivative (BatchInputType const &pattern, BatchOutputType const &outputs, BatchOutputType const &coefficients, State const &state, RealVector &gradient) const
	calculates the weighted sum of derivatives w.r.t the parameters.

SHARK_EXPORT_SYMBOL void	setTrainingParameters (bool centers, bool width)
	Enables or disables parameters for learning.

BatchInputType const &	centers () const
	Returns the center values of the neurons.

BatchInputType &	centers ()
	Sets the center values of the neurons.

RealVector const &	gamma () const
	Returns the width parameter of the Gaussian functions.

SHARK_EXPORT_SYMBOL void	setGamma (RealVector const &gamma)
	sets the width parameters - the gamma values - of the distributions.

SHARK_EXPORT_SYMBOL void	read (InArchive &archive)
	From ISerializable, reads a model from an archive.

SHARK_EXPORT_SYMBOL void	write (OutArchive &archive) const
	From ISerializable, writes a model to an archive.

Public Member Functions inherited from shark::AbstractModel< RealVector, RealVector >
	AbstractModel ()

virtual	~AbstractModel ()

const Features &	features () const

virtual void	updateFeatures ()

bool	hasFirstParameterDerivative () const
	Returns true when the first parameter derivative is implemented.

bool	hasFirstInputDerivative () const
	Returns true when the first input derivative is implemented.

virtual void	eval (BatchInputType const &patterns, BatchOutputType &outputs) const
	Standard interface for evaluating the response of the model to a batch of patterns.

virtual void	eval (InputType const &pattern, OutputType &output) const
	Standard interface for evaluating the response of the model to a single pattern.

Data< OutputType >	operator() (Data< InputType > const &patterns) const
	Model evaluation as an operator for a whole dataset. This is a convenience function.

OutputType	operator() (InputType const &pattern) const
	Model evaluation as an operator for a single pattern. This is a convenience function.

BatchOutputType	operator() (BatchInputType const &patterns) const
	Model evaluation as an operator for a single pattern. This is a convenience function.

virtual void	weightedInputDerivative (BatchInputType const &pattern, BatchOutputType const &outputs, BatchOutputType const &coefficients, State const &state, BatchInputType &derivative) const
	calculates the weighted sum of derivatives w.r.t the inputs

virtual void	weightedDerivatives (BatchInputType const &patterns, BatchOutputType const &outputs, BatchOutputType const &coefficients, State const &state, RealVector &parameterDerivative, BatchInputType &inputDerivative) const
	calculates weighted input and parameter derivative at the same time

Public Member Functions inherited from shark::IParameterizable< VectorType >
virtual	~IParameterizable ()

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

Public Member Functions inherited from shark::ISerializable
virtual	~ISerializable ()
	Virtual d'tor.

void	load (InArchive &archive, unsigned int version)
	Versioned loading of components, calls read(...).

void	save (OutArchive &archive, unsigned int version) const
	Versioned storing of components, calls write(...).

	BOOST_SERIALIZATION_SPLIT_MEMBER ()

Protected Attributes
RealMatrix	m_centers
	The center points. The i-th element corresponds to the center of neuron number i.

RealVector	m_gamma
	stores the width parameters of the Gaussian functions

RealVector	m_logNormalization
	the logarithm of the normalization constant for every distribution

bool	m_trainCenters
	enables learning of the center points of the neurons

bool	m_trainWidth
	enables learning of the width parameters.

Protected Attributes inherited from shark::AbstractModel< RealVector, RealVector >
Features	m_features

Additional Inherited Members
Public Types inherited from shark::AbstractModel< RealVector, RealVector >
enum	Feature

typedef RealVector	InputType
	Defines the input type of the model.

typedef RealVector	OutputType
	Defines the output type of the model.

typedef AbstractModel< RealVector, RealVector, RealVector >	ModelBaseType
	Defines the BaseType used by the model (this type). Useful for creating derived models.

typedef Batch< InputType >::type	BatchInputType
	defines the batch type of the input type.

typedef Batch< OutputType >::type	BatchOutputType
	defines the batch type of the output type

typedef TypedFlags< Feature >	Features

typedef TypedFeatureNotAvailableException< Feature >	FeatureNotAvailableException

Public Types inherited from shark::IParameterizable< VectorType >
typedef VectorType	ParameterVectorType

Detailed Description

Implements a layer of radial basis functions in a neural network.

A Radial basis function layer as modeled in shark is a set of N Gaussian distributions \( p(x|i) \).

\[ p(x|i) = e^{\gamma_i*\|x-m_i\|^2} \]

and the layer transforms an input x to a vector \((p(x|1),\dots,p(x|N)\). The \(\gamma_i\) govern the width of the Gaussians, while the vectors \( m_i \) set the centers of every Gaussian distribution.

RBF networks profit much from good guesses on the centers and kernel function parameters. In case of a Gaussian kernel a call to k-Means or the EM-algorithm can be used to get a good initialisation for the network.

Definition at line 57 of file RBFLayer.h.

Constructor & Destructor Documentation

◆ RBFLayer() [1/2]

SHARK_EXPORT_SYMBOL shark::RBFLayer::RBFLayer ( )

Creates an empty Radial Basis Function layer.

◆ RBFLayer() [2/2]

SHARK_EXPORT_SYMBOL shark::RBFLayer::RBFLayer	(	std::size_t	numInput,
		std::size_t	numOutput
	)

Creates a layer of a Radial Basis Function Network.

This method creates a Radial Basis Function Network (RBFN) with numInput input neurons and numOutput output neurons.

Parameters

numInput	Number of input neurons, equal to dimensionality of input space.
numOutput	Number of output neurons, equal to dimensionality of output space and number of gaussian distributions

Member Function Documentation

◆ centers() [1/2]

BatchInputType & shark::RBFLayer::centers ( )

inline

Sets the center values of the neurons.

Definition at line 147 of file RBFLayer.h.

References m_centers.

◆ centers() [2/2]

BatchInputType const & shark::RBFLayer::centers ( ) const

inline

Returns the center values of the neurons.

Definition at line 143 of file RBFLayer.h.

References m_centers.

◆ createState()

boost::shared_ptr< State > shark::RBFLayer::createState ( ) const

inlinevirtual

Creates an internal state of the model.

The state is needed when the derivatives are to be calculated. Eval can store a state which is then reused to speed up the calculations of the derivatives. This also allows eval to be evaluated in parallel!

Reimplemented from shark::AbstractModel< RealVector, RealVector >.

Definition at line 109 of file RBFLayer.h.

◆ eval()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::eval	(	BatchInputType const &	patterns,
		BatchOutputType &	outputs,
		State &	state
	)		const

virtual

Standard interface for evaluating the response of the model to a batch of patterns.

Parameters

patterns	the inputs of the model
outputs	the predictions or response of the model to every pattern
state	intermediate results stored by eval which can be reused for derivative computation.

Implements shark::AbstractModel< RealVector, RealVector >.

◆ gamma()

RealVector const & shark::RBFLayer::gamma ( ) const

inline

Returns the width parameter of the Gaussian functions.

Definition at line 152 of file RBFLayer.h.

References m_gamma.

◆ inputShape()

Shape shark::RBFLayer::inputShape ( ) const

inlinevirtual

Returns the number of input neurons.

Implements shark::AbstractModel< RealVector, RealVector >.

Definition at line 100 of file RBFLayer.h.

References m_centers.

◆ name()

std::string shark::RBFLayer::name ( ) const

inlinevirtual

From INameable: return the class name.

Reimplemented from shark::INameable.

Definition at line 84 of file RBFLayer.h.

◆ numberOfParameters()

SHARK_EXPORT_SYMBOL std::size_t shark::RBFLayer::numberOfParameters ( ) const

virtual

Returns the number of parameters which are currently enabled for training.

Reimplemented from shark::IParameterizable< VectorType >.

◆ outputShape()

Shape shark::RBFLayer::outputShape ( ) const

inlinevirtual

Returns the number of output neurons.

Implements shark::AbstractModel< RealVector, RealVector >.

Definition at line 105 of file RBFLayer.h.

References m_centers.

◆ parameterVector()

SHARK_EXPORT_SYMBOL RealVector shark::RBFLayer::parameterVector ( ) const

virtual

Returns the current parameter vector. The amount and order of weights depend on the training parameters.

The format of the parameter vector is \( (m_1,\dots,m_k,\log(\gamma_1),\dots,\log(\gamma_k))\) if training of one or more parameters is deactivated, they are removed from the parameter vector

Reimplemented from shark::IParameterizable< VectorType >.

◆ read()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::read ( InArchive & archive )

virtual

From ISerializable, reads a model from an archive.

Reimplemented from shark::AbstractModel< RealVector, RealVector >.

◆ setGamma()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::setGamma ( RealVector const & gamma )

sets the width parameters - the gamma values - of the distributions.

◆ setParameterVector()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::setParameterVector ( RealVector const & newParameters )

virtual

Sets the new internal parameters.

Reimplemented from shark::IParameterizable< VectorType >.

◆ setStructure()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::setStructure	(	std::size_t	numInput,
		std::size_t	numOutput
	)

Configures a Radial Basis Function Network.

This method initializes the structure of the Radial Basis Function Network (RBFN) with numInput input neurons, numOutput output neurons and numHidden hidden neurons.

Parameters

numInput	Number of input neurons, equal to dimensionality of input space.
numOutput	Number of output neurons (basis functions), equal to dimensionality of output space.

◆ setTrainingParameters()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::setTrainingParameters	(	bool	centers,
		bool	width
	)

Enables or disables parameters for learning.

Parameters

centers	whether the centers should be trained
width	whether the distribution width should be trained

◆ weightedParameterDerivative()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::weightedParameterDerivative	(	BatchInputType const &	pattern,
		BatchOutputType const &	outputs,
		BatchOutputType const &	coefficients,
		State const &	state,
		RealVector &	derivative
	)		const

virtual

calculates the weighted sum of derivatives w.r.t the parameters.

Parameters

pattern	the patterns to evaluate
outputs	the target outputs
coefficients	the coefficients which are used to calculate the weighted sum for every pattern
state	intermediate results stored by eval to speed up calculations of the derivatives
derivative	the calculated derivative as sum over all derivates of all patterns

Reimplemented from shark::AbstractModel< RealVector, RealVector >.

◆ write()

SHARK_EXPORT_SYMBOL void shark::RBFLayer::write ( OutArchive & archive ) const

virtual

From ISerializable, writes a model to an archive.

Reimplemented from shark::AbstractModel< RealVector, RealVector >.

Member Data Documentation

◆ m_centers

RealMatrix shark::RBFLayer::m_centers

protected

The center points. The i-th element corresponds to the center of neuron number i.

Definition at line 168 of file RBFLayer.h.

Referenced by centers(), centers(), inputShape(), and outputShape().

◆ m_gamma

RealVector shark::RBFLayer::m_gamma

protected

stores the width parameters of the Gaussian functions

Definition at line 171 of file RBFLayer.h.

Referenced by gamma().

◆ m_logNormalization

RealVector shark::RBFLayer::m_logNormalization

protected

the logarithm of the normalization constant for every distribution

Definition at line 174 of file RBFLayer.h.

◆ m_trainCenters

bool shark::RBFLayer::m_trainCenters

protected

enables learning of the center points of the neurons

Definition at line 178 of file RBFLayer.h.

◆ m_trainWidth

bool shark::RBFLayer::m_trainWidth

protected

enables learning of the width parameters.

Definition at line 180 of file RBFLayer.h.

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

include/shark/Models/RBFLayer.h

Public Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ RBFLayer() [1/2]

◆ RBFLayer() [2/2]

Member Function Documentation

◆ centers() [1/2]

◆ centers() [2/2]

◆ createState()

◆ eval()

◆ gamma()

◆ inputShape()

◆ name()

◆ numberOfParameters()

◆ outputShape()

◆ parameterVector()

◆ read()

◆ setGamma()

◆ setParameterVector()

◆ setStructure()

◆ setTrainingParameters()

◆ weightedParameterDerivative()

◆ write()

Member Data Documentation

◆ m_centers

◆ m_gamma

◆ m_logNormalization

◆ m_trainCenters

◆ m_trainWidth