ConjugateGradientOptimiser Class Reference

#include <ConjugateGradientOptimiser.h>

Inheritance diagram for ConjugateGradientOptimiser:

Collaboration diagram for ConjugateGradientOptimiser:

Public Member Functions
	ConjugateGradientOptimiser ()
	~ConjugateGradientOptimiser ()
void	Optimise (PrecisionType *startingPoint)
virtual void	Initialise ()
void	SetContinueOptimisation (bool continueOptimisationIn)
Public Member Functions inherited from Optimiser
	Optimiser ()
virtual	~Optimiser ()
void	SetObjectiveFunction (std::shared_ptr< ObjectiveFunction > &objectiveFunctionIn)
virtual void	Optimise (PrecisionType *startingPoint)=0
virtual void	Initialise ()=0
PrecisionType	GetBestObjectiveFunctionValue ()
PrecisionType	GetInitialObjectiveFunctionValue ()
PrecisionType *	GetBestPoint ()
void	ResetCurrentIterationNumber ()
void	SetMaxIterations (unsigned int maxIterations)
void	SetPreviouslyEvaluatedObjectiveFunctionValue (PrecisionType objectiveFunctionValue)
void	SetOutputIntermediateGradientFolder (const std::string &outputFolderIn, const std::string &outPrefix)

Private Member Functions
void	CalculateGradientUpdate ()
void	SaveIntermediateGradientImage ()

Private Attributes
PrecisionType *	array1
	The array g in numerical recipes. More...
PrecisionType *	array2
	The array h in numerical recipes. More...
bool	continuationOfOptimisationPossible
	Indicating if array1 and array2 were initialised properly and can be used to continue with the optimisation. More...
bool	continueOptimisation
	Indicates if the sequence of gradients should be continued. Reg-resp implemented this to be set to true, but it may be beneficial to set this to false and start with a fresh gradient sequence if the motion-compensated image and thus the objective function changed with each switch iteration. More...

Additional Inherited Members
Public Types inherited from Optimiser
typedef float	PrecisionType
Protected Member Functions inherited from Optimiser
void	MoveCurrentPointAlongGradient (PrecisionType stepSize)
void	PerformLineSearch (PrecisionType maxLength, PrecisionType smallLength, PrecisionType &startLength)
Protected Attributes inherited from Optimiser
PrecisionType	currentObjectiveFunctionValue
	The current objective function value corresponding to the current iteration. More...
PrecisionType	bestObjectiveFunctionValue
	The best objective function value achieved. More...
PrecisionType	initialObjectiveFunctionValue
	The objective function value at the starting point of the optimisation. More...
unsigned int	currentIterationNumber
	Tracker of the current iteration number. More...
unsigned int	maxNumberOfIterations
	The maximum number of allowed iterations. More...
const unsigned int	maxNumberOfLineIterations
	The maximum number of iterations used for the line-serach along the gradient. More...
PrecisionType	maxLineSearchStepSize
	The maximum step size used for the line-search. Has to be initialised in derived class. More...
PrecisionType	minLineSearchStepSize
	The minimum step size used for the line-search. Has to be initialised in derived class. More...
unsigned int	numberOfIterationsPerformed
	The total number of iterations performed. More...
unsigned int	numberOfDOFs
	The number of degrees of freedom. More...
std::shared_ptr< ObjectiveFunction >	objectiveFunction
	Shared pointer to the objective function. More...
PrecisionType *	gradient
	The array xi in numerical recipes. More...
PrecisionType *	currentPoint
	The current point, p in numerical recipes. More...
PrecisionType *	bestPoint
	The best point achieved, p in numerical recipes. More...
PrecisionType	preEvaluatedObjectiveFunctionValue
	A pre-calculated objective function value that will be used during the optimisation if usePreEvaluatedEvaluatedObjectiveFunctionValue is set to true. More...
bool	usePreEvaluatedEvaluatedObjectiveFunctionValue
	Indicator if the first objective function value calculation can be skipped. More...
std::string	outputIntermediateGradientFolder
	Folder to which the intermediated objective function gradients will be saved. More...
std::string	outputIntermediateGradientPrefix
	String holding additional information to differentiate the output from different levels. More...

Detailed Description

Class implementing the conjugate gradient optimiser according to nifty-reg-resp and Numerical Recipes in C++.

Constructor & Destructor Documentation

ConjugateGradientOptimiser()

ConjugateGradientOptimiser::ConjugateGradientOptimiser

(

)

Constructor

~ConjugateGradientOptimiser()

ConjugateGradientOptimiser::~ConjugateGradientOptimiser

(

)

Destructor

Member Function Documentation

CalculateGradientUpdate()

void ConjugateGradientOptimiser::CalculateGradientUpdate ( )

private

Calculate the conjugate gradient update

Initialise()

void ConjugateGradientOptimiser::Initialise ( )

virtual

Initialise the optimiser. Initialises all parameters of the Optimiser base class and conjugate gradient optimiser.

Implements Optimiser.

Optimise()

void ConjugateGradientOptimiser::Optimise ( PrecisionType *  startingPoint )

virtual

Initiate the optimisation. The iteration numbers as well as the stored objective function values will be reset.

Parameters

startingPoint

Point where to start the optimisation from. Can be null-pointer to start form zero.

Implements Optimiser.

SaveIntermediateGradientImage()

void ConjugateGradientOptimiser::SaveIntermediateGradientImage ( )

private

Function that save the current gradient to nifti images if the output file name is not empty.

Note

This will trigger recalculation of the gradient and will affect performance.

SetContinueOptimisation()

void ConjugateGradientOptimiser::SetContinueOptimisation ( bool  continueOptimisationIn )

inline

Set if the gradient sequence should be continued even if the objective function changed. Reg-resp implemented this to be set to true, but it may be beneficial to set this to false and start with a fresh gradient sequence if the motion-compensated image and thus the objective function changed with each switch iteration.

Parameters

continueOptimisationIn

Define the continuation of gradient calculation. Set to false for a fresh start.

Member Data Documentation

array1

PrecisionType* ConjugateGradientOptimiser::array1

private

The array g in numerical recipes.

array2

PrecisionType* ConjugateGradientOptimiser::array2

private

The array h in numerical recipes.

continuationOfOptimisationPossible

bool ConjugateGradientOptimiser::continuationOfOptimisationPossible

private

Indicating if array1 and array2 were initialised properly and can be used to continue with the optimisation.

continueOptimisation

bool ConjugateGradientOptimiser::continueOptimisation

private

Indicates if the sequence of gradients should be continued. Reg-resp implemented this to be set to true, but it may be beneficial to set this to false and start with a fresh gradient sequence if the motion-compensated image and thus the objective function changed with each switch iteration.

---

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

• motion-lib/ConjugateGradientOptimiser.h
• motion-lib/ConjugateGradientOptimiser.cpp