LBFGS

Classes
class	turi::optimization::lbfgs_solver

Functions
	turi::optimization::lbfgs_solver::lbfgs_solver (std::shared_ptr< first_order_opt_interface > _model)
void	turi::optimization::lbfgs_solver::setup (const DenseVector &init_point, const std::map< std::string, flexible_type > &opts, const std::shared_ptr< smooth_regularizer_interface > &reg=nullptr)
bool	turi::optimization::lbfgs_solver::next_iteration ()
const solver_status &	turi::optimization::lbfgs_solver::status () const
solver_status	turi::optimization::lbfgs (std::shared_ptr< first_order_opt_interface > model, const DenseVector &init_point, const std::map< std::string, flexible_type > &opts, const std::shared_ptr< smooth_regularizer_interface > &reg=nullptr)

Variables
size_t	turi::optimization::solver_status::iteration = 0
double	turi::optimization::solver_status::solver_time = 0
DenseVector	turi::optimization::solver_status::solution
DenseVector	turi::optimization::solver_status::gradient
DenseMatrix	turi::optimization::solver_status::hessian
double	turi::optimization::solver_status::residual = NAN
double	turi::optimization::solver_status::function_value = NAN
size_t	turi::optimization::solver_status::num_function_evaluations = 0
size_t	turi::optimization::solver_status::num_gradient_evaluations = 0
double	turi::optimization::solver_status::step_size = 0
OPTIMIZATION_STATUS	turi::optimization::solver_status::status = OPTIMIZATION_STATUS::OPT_UNSET

Detailed Description

Function Documentation

lbfgs_solver()

turi::optimization::lbfgs_solver::lbfgs_solver ( std::shared_ptr< first_order_opt_interface >  _model )

inline

Construct the solver around a specific model interface.

Parameters

[in]

model

Model with first order optimization interface.

Definition at line 100 of file lbfgs.hpp.

lbfgs()

solver_status turi::optimization::lbfgs	(	std::shared_ptr< first_order_opt_interface >	model,
		const DenseVector &	init_point,
		const std::map< std::string, flexible_type > &	opts,
		const std::shared_ptr< smooth_regularizer_interface > &	reg = nullptr
	)

Solves lbgfgs problem end-to-end.

This class wraps the above iterative solver in a convenience function, iterating the solution until completion.

Parameters

model	The implementation of first_order_opt_interface used in the optimization.
init_point	The initial point at which the optimization starts.
opts	The options. Uses all the options given to setup() in the lbfgs_solver class, plus "max_iterations" to terminate the optimization after a given number of iterations.
reg	Optional regularization interface.

next_iteration()

bool turi::optimization::lbfgs_solver::next_iteration

(

)

Perform the next update of the solution.

Call this method repeatedly to perform the optimization. Each iteration updates the solution point with one step.

setup()

void turi::optimization::lbfgs_solver::setup	(	const DenseVector &	init_point,
		const std::map< std::string, flexible_type > &	opts,
		const std::shared_ptr< smooth_regularizer_interface > &	reg = nullptr
	)

Sets up (or resets) the solver.

Parameters

[in]	init_point	Starting point for the solver.
[in]	opts	Solver options. Options are "lbfgs_memory_level" and "convergence_threshold". If not given, defaults are taken from the table in optimization_interface.hpp.
[in]	reg	Shared ptr to an interface to a smooth regularizer.

status()

const solver_status& turi::optimization::lbfgs_solver::status ( ) const

inline

The status after a given iteration.

The best solution so far is given by status().solution.

Definition at line 129 of file lbfgs.hpp.

Variable Documentation

function_value

double turi::optimization::solver_status::function_value = NAN

Function value

Definition at line 36 of file lbfgs.hpp.

gradient

DenseVector turi::optimization::solver_status::gradient

Current gradient

Definition at line 33 of file lbfgs.hpp.

hessian

DenseMatrix turi::optimization::solver_status::hessian

Current hessian

Definition at line 34 of file lbfgs.hpp.

iteration

size_t turi::optimization::solver_status::iteration = 0

Iterations taken

Definition at line 30 of file lbfgs.hpp.

num_function_evaluations

size_t turi::optimization::solver_status::num_function_evaluations = 0

Function evals

Definition at line 37 of file lbfgs.hpp.

num_gradient_evaluations

size_t turi::optimization::solver_status::num_gradient_evaluations = 0

Gradient evals

Definition at line 38 of file lbfgs.hpp.

residual

double turi::optimization::solver_status::residual = NAN

Residual norm

Definition at line 35 of file lbfgs.hpp.

solution

DenseVector turi::optimization::solver_status::solution

Current Solution

Definition at line 32 of file lbfgs.hpp.

solver_time

double turi::optimization::solver_status::solver_time = 0

Wall clock time (s)

Definition at line 31 of file lbfgs.hpp.

status

OPTIMIZATION_STATUS turi::optimization::solver_status::status = OPTIMIZATION_STATUS::OPT_UNSET

Status

Definition at line 41 of file lbfgs.hpp.

step_size

double turi::optimization::solver_status::step_size = 0

Current step size

Definition at line 39 of file lbfgs.hpp.