Optimization Algorithms

Classes
class	npl::BFGSOpt
class	npl::GradientOpt
class	npl::LBFGSOpt
	Limited-Memory Broyden–Fletcher–Goldfarb–Shanno Algorithm based on "A Limited Memory Algorithm for Bound Constrained Optimization" Richard H. Byrd, Peihuang Lu, Jorge Nocedal, and Ciyou Zhu http://dx.doi.org/10.1137/0916069. More...
class	npl::Optimizer

Typedefs
typedef function< int(const VectorXd &x, double &v, VectorXd &g)>	npl::ValGradFunc
	Value and Gradient Computation Function. More...
typedef function< int(const VectorXd &x, VectorXd &g)>	npl::GradFunc
	Gradient Only Computation Function. More...
typedef function< int(const VectorXd &x, double &v)>	npl::ValFunc
	Value Only Computation Function. More...
typedef function< int(const VectorXd &x, double v, const VectorXd &g, size_t iter)>	npl::CallBackFunc
	Callback function. More...

Enumerations
enum	npl::StopReason {   npl::ENDGRAD, npl::ENDSTEP, npl::ENDVALUE, npl::ENDABSVALUE,   npl::ENDITERS, npl::ENDFAIL }

Functions
int	npl::testgrad (double &error, const VectorXd &x, double stepsize, double tol, const ValFunc &valfunc, const ValGradFunc &valgradfunc)
	Tests a gradient function using the value function. More...
int	npl::testgrad (double &error, const VectorXd &x, double stepsize, double tol, const ValFunc &valfunc, const GradFunc &gradfunc)
	Tests a gradient function using the value function. More...
int	npl::gRosenbrock_G (const VectorXd &x, VectorXd &gradient)
	Implements generized rosenbrock gradient. More...
int	npl::gRosenbrock_V (const VectorXd &x, double &v)
	Implements generized rosenbrock value. More...
void	npl::gRosenbrock_callCounts (size_t &vcalls, size_t &gcalls)
	Returns the number of times the Value and Gradient functions for the Generalized Rosenbrock Function were called. More...
int	npl::noopCallback (const VectorXd &x, double value, const VectorXd &grad, size_t iter)
	Callback that does nothing. More...

Detailed Description

Typedef Documentation

typedef function<int(const VectorXd& x, double v, const VectorXd& g, size_t iter)> npl::CallBackFunc

Callback function.

Definition at line 56 of file opt.h.

typedef function<int(const VectorXd& x, VectorXd& g)> npl::GradFunc

Gradient Only Computation Function.

Definition at line 46 of file opt.h.

typedef function<int(const VectorXd& x, double& v)> npl::ValFunc

Value Only Computation Function.

Definition at line 51 of file opt.h.

typedef function<int(const VectorXd& x, double& v, VectorXd& g)> npl::ValGradFunc

Value and Gradient Computation Function.

Definition at line 41 of file opt.h.

Enumeration Type Documentation

enum npl::StopReason

Enumerator
ENDGRAD
ENDSTEP
ENDVALUE
ENDABSVALUE
ENDITERS
ENDFAIL

Definition at line 141 of file opt.h.

Function Documentation

void npl::gRosenbrock_callCounts	(	size_t &	vcalls,
		size_t &	gcalls
	)

Returns the number of times the Value and Gradient functions for the Generalized Rosenbrock Function were called.

Parameters

vcalls	Value calls
gcalls	Gradient calls

int npl::gRosenbrock_G	(	const VectorXd &	x,
		VectorXd &	gradient
	)

Implements generized rosenbrock gradient.

Parameters

x	Position vector
gradient	Gradient at the position

Returns

int npl::gRosenbrock_V	(	const VectorXd &	x,
		double &	v
	)

Implements generized rosenbrock value.

Parameters

x	Position vector
v	values

Returns

int npl::noopCallback ( const VectorXd &  x, double  value, const VectorXd &  grad, size_t  iter  )

inline

Callback that does nothing.

Parameters

x
value
grad
iter

Returns

Definition at line 132 of file opt.h.

int npl::testgrad	(	double &	error,
		const VectorXd &	x,
		double	stepsize,
		double	tol,
		const ValFunc &	valfunc,
		const ValGradFunc &	valgradfunc
	)

Tests a gradient function using the value function.

Parameters

error	Error between analytical and numeric gradient
x	Position to test
stepsize	Step to take when testing gradient (will be taken in each dimension successively)
tol	Tolerance, error below the tolerance will cause the function to return 0, higher error will cause the function to return -1
valfunc	Function values compute
valgradfunc	Function gradient compute

Returns

int npl::testgrad	(	double &	error,
		const VectorXd &	x,
		double	stepsize,
		double	tol,
		const ValFunc &	valfunc,
		const GradFunc &	gradfunc
	)

Tests a gradient function using the value function.

Parameters

error	Error between analytical and numeric gradient
x	Position to test
stepsize	Step to take when testing gradient (will be taken in each dimension successively)
tol	Tolerance, error below the tolerance will cause the function to return 0, higher error will cause the function to return -1
valfunc	Function values compute
gradfunc	Function gradient compute

Returns