BIAS::LevenbergMarquardtBase Class Reference

Base interface for classes using non-linear optimization of a target function using the Levenberg-Marquardt algorithm. More...

#include <MathAlgo/LevenbergMarquardtBase.hh>

Inheritance diagram for BIAS::LevenbergMarquardtBase:

Public Member Functions
	LevenbergMarquardtBase ()
int	LM_Compute (const int numResiduals, const int numParameters, BIAS::Vector< double > &startX, BIAS::Vector< double > &resultX, BIAS::Vector< double > &residuals)
	Compute Levenberg-Marquardt algorithm using the user defined target function f(x) implemented in LM_TargetFunction_, and the analytic Jacobian J(x) implemented in LM_Jacobian_. More...
int	LM_ComputeWithoutJacobian (const int numResiduals, const int numParameters, BIAS::Vector< double > &startX, BIAS::Vector< double > &resultX, BIAS::Vector< double > &residuals, const double epsilon=LM_DEF_EPSILON)
	Compute Levenberg-Marquardt algorithm using the user defined target function f(x) implemented in LM_TargetFunction_. More...
BIAS::Matrix< double >	LM_GetJacobianResult () const
	Return the resulting Jacobian J(x) from the last computation. More...
int	LM_GetMaxIterations () const
	Return the maximal number of iterations to perform. More...
int	LM_GetNumOfIterations () const
	Return the number of iterations performed in last computation. More...
BIAS::Vector< double >	LM_GetParameterResult () const
	Return the resulting parameter vector x from last computation. More...
BIAS::Vector< double >	LM_GetResidualsResult () const
	Return the resulting residual vector f(x) from last computation. More...
double	LM_GetSumOfSquaredErrors () const
	Return the sum of squared errors from the last computation. More...
double	LM_GetTolerance () const
	Return tolerance used to evaluate convergence. More...
void	LM_SetMaxIterations (const int iters)
	Set the maximal number of iterations to perform. More...
void	LM_SetTolerance (const double tol)
	Set tolerance used to evaluate convergence. More...
virtual	~LevenbergMarquardtBase ()

Protected Member Functions
virtual int	LM_JacobianFunction_ (const BIAS::Vector< double > &x, BIAS::Matrix< double > &J)
	Compute Jacobian J(x) of target function f evaluated at vector x. More...
virtual int	LM_TargetFunction_ (const BIAS::Vector< double > &x, BIAS::Vector< double > &residuals)=0
	Compute residuals of target function f for input vector x. More...

Detailed Description

Base interface for classes using non-linear optimization of a target function using the Levenberg-Marquardt algorithm.

Note

Since this class used virtual function calls to evaluate the target function of the Levenber-Marquardt instance, and data is copied between BIAS data structure and arrays in every iteration, it is not well suited for high-performance demands! Use the direct Minpack interface from Minpack.hh instead when you are targeting at performance!

Provides a class interface to Minpack with convenient definition of target function f(x) and Jacobian J(x).

See Also

Minpack.hh

Author

esquivel

Date

05/2010

Examples:

ExampleLevenbergMarquardt.cpp.

Definition at line 56 of file LevenbergMarquardtBase.hh.

Constructor & Destructor Documentation

BIAS::LevenbergMarquardtBase::LevenbergMarquardtBase ( )

inline

Definition at line 76 of file LevenbergMarquardtBase.hh.

virtual BIAS::LevenbergMarquardtBase::~LevenbergMarquardtBase ( )

inlinevirtual

Definition at line 84 of file LevenbergMarquardtBase.hh.

Member Function Documentation

int LevenbergMarquardtBase::LM_Compute	(	const int	numResiduals,
		const int	numParameters,
		BIAS::Vector< double > &	startX,
		BIAS::Vector< double > &	resultX,
		BIAS::Vector< double > &	residuals
	)

Compute Levenberg-Marquardt algorithm using the user defined target function f(x) implemented in LM_TargetFunction_, and the analytic Jacobian J(x) implemented in LM_Jacobian_.

Returns

Returns value < 0 if computation failed, 0 for convergence, and > 0 otherwise (see LevenbergMarquardt in #Minpack.hh).

Definition at line 30 of file LevenbergMarquardtBase.cpp.

References BIAS::LevenbergMarquardtExtended(), TNT::Vector< T >::newsize(), and TNT::Matrix< T >::newsize().

int LevenbergMarquardtBase::LM_ComputeWithoutJacobian	(	const int	numResiduals,
		const int	numParameters,
		BIAS::Vector< double > &	startX,
		BIAS::Vector< double > &	resultX,
		BIAS::Vector< double > &	residuals,
		const double	epsilon = LM_DEF_EPSILON
	)

Compute Levenberg-Marquardt algorithm using the user defined target function f(x) implemented in LM_TargetFunction_.

   The Jacobian of f is approximated numerically using the given
   value epsilon.

Parameters

epsilon

is "an input variable used in determining a suitable step length for the forward-difference approximation" of the Jacobian of f (see documentation of minpack/fdjac2.c)

Returns

Returns value < 0 if computation failed, 0 for convergence, and > 0 otherwise (see LevenbergMarquardt in #Minpack.hh).

Definition at line 57 of file LevenbergMarquardtBase.cpp.

References BIAS::ComputeJacobian(), BIAS::LevenbergMarquardtExtended(), TNT::Vector< T >::newsize(), TNT::Matrix< T >::newsize(), and BIAS::Matrix< T >::SetZero().

BIAS::Matrix<double> BIAS::LevenbergMarquardtBase::LM_GetJacobianResult ( ) const

inline

Return the resulting Jacobian J(x) from the last computation.

Definition at line 139 of file LevenbergMarquardtBase.hh.

int BIAS::LevenbergMarquardtBase::LM_GetMaxIterations ( ) const

inline

Return the maximal number of iterations to perform.

Definition at line 119 of file LevenbergMarquardtBase.hh.

int BIAS::LevenbergMarquardtBase::LM_GetNumOfIterations ( ) const

inline

Return the number of iterations performed in last computation.

Definition at line 127 of file LevenbergMarquardtBase.hh.

BIAS::Vector<double> BIAS::LevenbergMarquardtBase::LM_GetParameterResult ( ) const

inline

Return the resulting parameter vector x from last computation.

Definition at line 131 of file LevenbergMarquardtBase.hh.

BIAS::Vector<double> BIAS::LevenbergMarquardtBase::LM_GetResidualsResult ( ) const

inline

Return the resulting residual vector f(x) from last computation.

Definition at line 135 of file LevenbergMarquardtBase.hh.

double BIAS::LevenbergMarquardtBase::LM_GetSumOfSquaredErrors ( ) const

inline

Return the sum of squared errors from the last computation.

Definition at line 123 of file LevenbergMarquardtBase.hh.

double BIAS::LevenbergMarquardtBase::LM_GetTolerance ( ) const

inline

Return tolerance used to evaluate convergence.

Definition at line 111 of file LevenbergMarquardtBase.hh.

virtual int BIAS::LevenbergMarquardtBase::LM_JacobianFunction_ ( const BIAS::Vector< double > &  x, BIAS::Matrix< double > &  J  )

inlineprotectedvirtual

Compute Jacobian J(x) of target function f evaluated at vector x.

Note

Override this with the user specified Jacobian function. Otherwise, only a numerical approximation of J(x) will be used.

Definition at line 96 of file LevenbergMarquardtBase.hh.

void BIAS::LevenbergMarquardtBase::LM_SetMaxIterations ( const int  iters )

inline

Set the maximal number of iterations to perform.

Definition at line 115 of file LevenbergMarquardtBase.hh.

void BIAS::LevenbergMarquardtBase::LM_SetTolerance ( const double  tol )

inline

Set tolerance used to evaluate convergence.

Definition at line 107 of file LevenbergMarquardtBase.hh.

virtual int BIAS::LevenbergMarquardtBase::LM_TargetFunction_ ( const BIAS::Vector< double > &  x, BIAS::Vector< double > &  residuals  )

protectedpure virtual

Compute residuals of target function f for input vector x.

Note

Override this with the user specified target function.

---

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

• MathAlgo/LevenbergMarquardtBase.hh
• MathAlgo/LevenbergMarquardtBase.cpp