itk::SPSAOptimizer Class Reference
[Optimizers]

An optimizer based on simultaneous perturbation... More...

#include <itkSPSAOptimizer.h>

Inheritance diagram for itk::SPSAOptimizer:

[legend]Collaboration diagram for itk::SPSAOptimizer:

[legend]List of all members.

Public Types
typedef SPSAOptimizer	Self
typedef SingleValuedNonLinearOptimizer	Superclass
typedef SmartPointer< Self >	Pointer
typedef SmartPointer< const Self >	ConstPointer
enum	StopConditionType { Unknown, MaximumNumberOfIterations, BelowTolerance, MetricError }
Public Member Functions
virtual const char *	GetNameOfClass () const
virtual void	AdvanceOneStep (void)
void	StartOptimization (void)
void	ResumeOptimization (void)
void	StopOptimization (void)
virtual MeasureType	GetValue (void) const
virtual MeasureType	GetValue (const ParametersType &parameters) const
virtual void	GuessParameters (unsigned long numberOfGradientEstimates, double initialStepSize)
virtual unsigned long	GetCurrentIteration () const
virtual StopConditionType	GetStopCondition () const
virtual double	GetLearningRate () const
virtual double	GetGradientMagnitude () const
virtual const DerivativeType &	GetGradient ()
virtual double	GetStateOfConvergence () const
virtual void	Seta (double _arg)
virtual double	Geta () const
virtual void	Setc (double _arg)
virtual double	Getc () const
virtual void	SetA (double _arg)
virtual double	GetA () const
virtual void	SetAlpha (double _arg)
virtual double	GetAlpha () const
virtual void	SetGamma (double _arg)
virtual double	GetGamma () const
virtual bool	GetMaximize () const
virtual void	SetMaximize (bool _arg)
virtual void	MaximizeOn ()
virtual void	MaximizeOff ()
bool	GetMinimize () const
void	SetMinimize (bool v)
void	MinimizeOn ()
void	MinimizeOff ()
virtual void	SetNumberOfPerturbations (unsigned long _arg)
virtual unsigned long	GetNumberOfPerturbations () const
virtual void	SetStateOfConvergenceDecayRate (double _arg)
virtual double	GetStateOfConvergenceDecayRate () const
virtual void	SetMinimumNumberOfIterations (unsigned long _arg)
virtual unsigned long	GetMinimumNumberOfIterations () const
virtual void	SetMaximumNumberOfIterations (unsigned long _arg)
virtual unsigned long	GetMaximumNumberOfIterations () const
virtual void	SetTolerance (double _arg)
virtual double	GetTolerance () const
Static Public Member Functions
Pointer	New ()
Protected Member Functions
	SPSAOptimizer ()
virtual	~SPSAOptimizer ()
void	PrintSelf (std::ostream &os, Indent indent) const
virtual double	Compute_a (unsigned long k) const
virtual double	Compute_c (unsigned long k) const
virtual void	GenerateDelta (const unsigned int spaceDimension)
virtual void	ComputeGradient (const ParametersType &parameters, DerivativeType &gradient)
Protected Attributes
DerivativeType	m_Gradient
double	m_LearningRate
DerivativeType	m_Delta
bool	m_Stop
StopConditionType	m_StopCondition
double	m_StateOfConvergence
unsigned long	m_CurrentIteration

---

Detailed Description

An optimizer based on simultaneous perturbation...

This optimizer is an implementation of the Simultaneous Perturbation Stochastic Approximation method, described in:

• http://www.jhuapl.edu/SPSA/

• Spall, J.C. (1998), "An Overview of the Simultaneous Perturbation Method for Efficient Optimization," Johns Hopkins APL Technical Digest, vol. 19, pp. 482-492

Definition at line 26 of file itkSPSAOptimizer.h.

---

Member Typedef Documentation

typedef SmartPointer<const Self> itk::SPSAOptimizer::ConstPointer

Reimplemented from itk::SingleValuedNonLinearOptimizer. Definition at line 35 of file itkSPSAOptimizer.h.

typedef SmartPointer<Self> itk::SPSAOptimizer::Pointer

Reimplemented from itk::SingleValuedNonLinearOptimizer. Definition at line 34 of file itkSPSAOptimizer.h.

typedef SPSAOptimizer itk::SPSAOptimizer::Self

Standard class typedefs. Reimplemented from itk::SingleValuedNonLinearOptimizer. Definition at line 32 of file itkSPSAOptimizer.h.

typedef SingleValuedNonLinearOptimizer itk::SPSAOptimizer::Superclass

Reimplemented from itk::SingleValuedNonLinearOptimizer. Definition at line 33 of file itkSPSAOptimizer.h.

---

Member Enumeration Documentation

enum itk::SPSAOptimizer::StopConditionType

Codes of stopping conditions Enumeration values: Unknown  MaximumNumberOfIterations  BelowTolerance  MetricError  Definition at line 44 of file itkSPSAOptimizer.h.

---

Constructor & Destructor Documentation

itk::SPSAOptimizer::SPSAOptimizer (   )  [protected]

virtual itk::SPSAOptimizer::~SPSAOptimizer (   )  [inline, protected, virtual]

Definition at line 187 of file itkSPSAOptimizer.h. References m_CurrentIteration, and m_StateOfConvergence.

---

Member Function Documentation

virtual void itk::SPSAOptimizer::AdvanceOneStep (  void   )  [virtual]

Advance one step following the gradient direction.

virtual double itk::SPSAOptimizer::Compute_a (  unsigned long  k  )  const [protected, virtual]

Method to compute the learning rate at iteration k (a_k).

virtual double itk::SPSAOptimizer::Compute_c (  unsigned long  k  )  const [protected, virtual]

Method to compute the gain factor for the perturbation at iteration k (c_k).

virtual void itk::SPSAOptimizer::ComputeGradient (  const ParametersType &  parameters, DerivativeType &  gradient )  [protected, virtual]

Compute the gradient at a position. m_NumberOfPerturbations are used, and scales are taken into account.

virtual void itk::SPSAOptimizer::GenerateDelta (  const unsigned int  spaceDimension  )  [protected, virtual]

Method to generate a perturbation vector. Takes scales into account.

virtual double itk::SPSAOptimizer::GetA (   )  const [virtual]

Set/Get A.

virtual double itk::SPSAOptimizer::Geta (   )  const [virtual]

Set/Get a.

virtual double itk::SPSAOptimizer::GetAlpha (   )  const [virtual]

Set/Get alpha.

virtual double itk::SPSAOptimizer::Getc (   )  const [virtual]

Set/Get c.

virtual unsigned long itk::SPSAOptimizer::GetCurrentIteration (   )  const [virtual]

Get the current iteration number.

virtual double itk::SPSAOptimizer::GetGamma (   )  const [virtual]

Set/Get gamma.

virtual const DerivativeType& itk::SPSAOptimizer::GetGradient (   )  [virtual]

Get the latest computed gradient

virtual double itk::SPSAOptimizer::GetGradientMagnitude (   )  const [virtual]

Get the GradientMagnitude of the latest computed gradient

virtual double itk::SPSAOptimizer::GetLearningRate (   )  const [virtual]

Get the current LearningRate (a_k)

virtual bool itk::SPSAOptimizer::GetMaximize (   )  const [virtual]

Methods to configure the cost function.

virtual unsigned long itk::SPSAOptimizer::GetMaximumNumberOfIterations (   )  const [virtual]

Set/Get the maximum number of iterations.

bool itk::SPSAOptimizer::GetMinimize (   )  const [inline]

Methods to configure the cost function. Definition at line 131 of file itkSPSAOptimizer.h. References MaximizeOff().

virtual unsigned long itk::SPSAOptimizer::GetMinimumNumberOfIterations (   )  const [virtual]

Set/Get the minimum number of iterations

virtual const char* itk::SPSAOptimizer::GetNameOfClass (   )  const [virtual]

Run-time type information (and related methods). Reimplemented from itk::SingleValuedNonLinearOptimizer.

virtual unsigned long itk::SPSAOptimizer::GetNumberOfPerturbations (   )  const [virtual]

Set/Get the number of perturbation used to construct a gradient estimate g_k. q = NumberOfPerturbations g_k = 1/q sum_{j=1..q} g^(j)_k

virtual double itk::SPSAOptimizer::GetStateOfConvergence (   )  const [virtual]

Get the state of convergence in the last iteration. When the StateOfConvergence is lower than the Tolerance, and the minimum number of iterations has been performed, the optimization stops. The state of convergence (SOC) is initialized with 0.0 and updated after each iteration as follows: SOC *= SOCDecayRate SOC += a_k * GradientMagnitude

virtual double itk::SPSAOptimizer::GetStateOfConvergenceDecayRate (   )  const [virtual]

Set/Get StateOfConvergenceDecayRate (number between 0 and 1).

virtual StopConditionType itk::SPSAOptimizer::GetStopCondition (   )  const [virtual]

Get Stop condition.

virtual double itk::SPSAOptimizer::GetTolerance (   )  const [virtual]

Set/Get Tolerance

virtual MeasureType itk::SPSAOptimizer::GetValue (  const ParametersType &  parameters  )  const [virtual]

Get the cost function value at any position Reimplemented from itk::SingleValuedNonLinearOptimizer.

virtual MeasureType itk::SPSAOptimizer::GetValue (  void   )  const [virtual]

Get the cost function value at the current position.

virtual void itk::SPSAOptimizer::GuessParameters (  unsigned long  numberOfGradientEstimates, double  initialStepSize )  [virtual]

Guess the parameters a and A. This function needs the number of GradientEstimates used for estimating a and A and and the expected initial step size (where step size is defined as the maximum of the absolute values of the parameter update). Make sure you set c, Alpha, Gamma, the MaximumNumberOfIterations, the Scales, and the the InitialPosition before calling this method. Described in: Spall, J.C. (1998), "Implementation of the Simultaneous Perturbation Algorithm for Stochastic Optimization", IEEE Trans. Aerosp. Electron. Syst. 34(3), 817-823.

virtual void itk::SPSAOptimizer::MaximizeOff (   )  [virtual]

Methods to configure the cost function. Referenced by GetMinimize().

virtual void itk::SPSAOptimizer::MaximizeOn (   )  [virtual]

Methods to configure the cost function. Referenced by SetMinimize().

void itk::SPSAOptimizer::MinimizeOff (  void   )  [inline]

Methods to configure the cost function. Definition at line 137 of file itkSPSAOptimizer.h.

void itk::SPSAOptimizer::MinimizeOn (  void   )  [inline]

Methods to configure the cost function. Definition at line 135 of file itkSPSAOptimizer.h.

Pointer itk::SPSAOptimizer::New (   )  [static]

Method for creation through the object factory. Reimplemented from itk::SingleValuedNonLinearOptimizer.

void itk::SPSAOptimizer::PrintSelf (  std::ostream &  os, Indent  indent )  const [protected, virtual]

PrintSelf method. Reimplemented from itk::SingleValuedNonLinearOptimizer.

void itk::SPSAOptimizer::ResumeOptimization (  void   )

Resume previously stopped optimization with current parameters See also: StopOptimization.

virtual void itk::SPSAOptimizer::SetA (  double  _arg  )  [virtual]

Set/Get A.

virtual void itk::SPSAOptimizer::Seta (  double  _arg  )  [virtual]

Set/Get a.

virtual void itk::SPSAOptimizer::SetAlpha (  double  _arg  )  [virtual]

Set/Get alpha.

virtual void itk::SPSAOptimizer::Setc (  double  _arg  )  [virtual]

Set/Get c.

virtual void itk::SPSAOptimizer::SetGamma (  double  _arg  )  [virtual]

Set/Get gamma.

virtual void itk::SPSAOptimizer::SetMaximize (  bool  _arg  )  [virtual]

Methods to configure the cost function.

virtual void itk::SPSAOptimizer::SetMaximumNumberOfIterations (  unsigned long  _arg  )  [virtual]

Set/Get the maximum number of iterations.

void itk::SPSAOptimizer::SetMinimize (  bool  v  )  [inline]

Methods to configure the cost function. Definition at line 133 of file itkSPSAOptimizer.h. References MaximizeOn().

virtual void itk::SPSAOptimizer::SetMinimumNumberOfIterations (  unsigned long  _arg  )  [virtual]

Set/Get the minimum number of iterations

virtual void itk::SPSAOptimizer::SetNumberOfPerturbations (  unsigned long  _arg  )  [virtual]

Set/Get the number of perturbation used to construct a gradient estimate g_k. q = NumberOfPerturbations g_k = 1/q sum_{j=1..q} g^(j)_k

virtual void itk::SPSAOptimizer::SetStateOfConvergenceDecayRate (  double  _arg  )  [virtual]

Set/Get StateOfConvergenceDecayRate (number between 0 and 1).

virtual void itk::SPSAOptimizer::SetTolerance (  double  _arg  )  [virtual]

Set/Get Tolerance

void itk::SPSAOptimizer::StartOptimization (  void   )  [virtual]

Start optimization. Reimplemented from itk::Optimizer.

void itk::SPSAOptimizer::StopOptimization (  void   )

Stop optimization. See also: ResumeOptimization

---

Member Data Documentation

unsigned long itk::SPSAOptimizer::m_CurrentIteration [protected]

Definition at line 199 of file itkSPSAOptimizer.h. Referenced by ~SPSAOptimizer().

DerivativeType itk::SPSAOptimizer::m_Delta [protected]

Definition at line 195 of file itkSPSAOptimizer.h.

DerivativeType itk::SPSAOptimizer::m_Gradient [protected]

Variables updated during optimization Definition at line 193 of file itkSPSAOptimizer.h.

double itk::SPSAOptimizer::m_LearningRate [protected]

Definition at line 194 of file itkSPSAOptimizer.h.

double itk::SPSAOptimizer::m_StateOfConvergence [protected]

Definition at line 198 of file itkSPSAOptimizer.h. Referenced by ~SPSAOptimizer().

bool itk::SPSAOptimizer::m_Stop [protected]

Definition at line 196 of file itkSPSAOptimizer.h.

StopConditionType itk::SPSAOptimizer::m_StopCondition [protected]

Definition at line 197 of file itkSPSAOptimizer.h.

---

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

• itkSPSAOptimizer.h

---