Detailed Description

Levenberg-Marquardt optimizer.

Public Member Functions
	LevenbergMarquardt ()
	Constructor The Levenberg Marquardt algorithm tries to minimize the sum of squares of a cost function with m_numMeasurements values over m_numParameters parameters.

void	init (int numParameters, int numMeasurements)
	Initialize buffers used in Levenberg Marquardt algorithm.

template<typename FuncEps, typename T>
int	runWithFiniteDifferences (Eigen::VectorXd &p, const T &x, const FuncEps &computeEps, double finiteDifferenceStepSize, int maxIter, double deltaTol, double fTol, Progress *progress=nullptr, double lambda=0.01)
	Run Levenberg Marquard using a finite-difference approximation of the Jacobian.

template<typename FuncEps, typename FuncJ, typename T>
int	run (Eigen::VectorXd &p, const T &x, const FuncEps &computeEps, const FuncJ &computeJ, int maxIter, double deltaTol, double fTol, Progress *progress=nullptr, double lambda=0.01)
	Run Levenberg Marquardt.

Member Function Documentation

void init	(	int	numParameters,
		int	numMeasurements )

inline

Initialize buffers used in Levenberg Marquardt algorithm.

Parameters

numParameters	number of parameters
numMeasurements	number of measurements The Levenberg Marquardt algorithm tries to minimize the sum of squares of a cost function with m_numMeasurements values over m_numParameters parameters.

template<typename FuncEps, typename T>

int runWithFiniteDifferences	(	Eigen::VectorXd &	p,
		const T &	x,
		const FuncEps &	computeEps,
		double	finiteDifferenceStepSize,
		int	maxIter,
		double	deltaTol,
		double	fTol,
		Progress *	progress = nullptr,
		double	lambda = 0.01 )

inline

Run Levenberg Marquard using a finite-difference approximation of the Jacobian.

Parameters

p	Initial parameter estimate. Is Replaced by optimized parameters by the algorithm.
x	auxiliary parameters. These are not directly used in the algorithm, but are passed to computeEps and computeJ as the second argument.
computeEps	Cost function with signature: void computeEps(const Eigen::Matrix<double,numParameters,1> &p, const T &x, Eigen::VectorXd &eps)
finiteDifferenceStepSize	Step size used for finite differences computing the Jacobian
maxIter	Maximum number of iterations
deltaTol	Update norm tolerance
fTol	Function tolerance
progress	Optional progress object for algorithm progress output.
lambda	Initial damping factor used by algorithm, defaults to 0.01 return >= 0 if successful, -1 otherwise. 0 = delta error below tolerance, 1 = function error below tolerance, 2 = max iterations reached, 3 = cancelled by user

template<typename FuncEps, typename FuncJ, typename T>

inline

Run Levenberg Marquardt.

Parameters

p	Initial parameter estimate. Is Replaced by optimized parameters by the algorithm.
x	auxiliary parameters. These are not directly used in the algorithm, but are passed to computeEps and computeJ as the second argument.
computeEps	Cost function with signature: void computeEps(const Eigen::Matrix<double,numParameters,1> &p, const T &x, Eigen::VectorXd &eps)
computeJ	Function with signature void computeJ(const Eigen::Matrix<double,numParameters,1> &p, const T &x, Eigen::Matrix<double,numMeasurements,numParameters> &J) for computing the Jacobian
maxIter	Maximum number of iterations
deltaTol	Update norm tolerance
fTol	Function tolerance
progress	Optional progress object for algorithm progress output.
lambda	Initial damping factor used by algorithm, defaults to 0.01 return >= 0 if successful, -1 otherwise. 0 = delta error below tolerance, 1 = function error below tolerance, 2 = max iterations reached, 3 = cancelled by user

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