lbfgs_parameter_t Struct Reference
[libLBFGS API]


Detailed Description

L-BFGS optimization parameters.

Call lbfgs_parameter_init() function to initialize parameters to the default values.


Data Fields

int m
 The number of corrections to approximate the inverse hessian matrix.
lbfgsfloatval_t epsilon
 Epsilon for convergence test.
int past
lbfgsfloatval_t delta
int max_iterations
 The maximum number of iterations.
int linesearch
 The line search algorithm.
int max_linesearch
 The maximum number of trials for the line search.
lbfgsfloatval_t min_step
 The minimum step of the line search routine.
lbfgsfloatval_t max_step
 The maximum step of the line search.
lbfgsfloatval_t ftol
 A parameter to control the accuracy of the line search routine.
lbfgsfloatval_t gtol
 A parameter to control the accuracy of the line search routine.
lbfgsfloatval_t xtol
 The machine precision for floating-point values.
lbfgsfloatval_t orthantwise_c
 Coeefficient for the L1 norm of variables.
int orthantwise_start
 Start index for computing L1 norm of the variables.
int orthantwise_end
 End index for computing L1 norm of the variables.


Field Documentation

int lbfgs_parameter_t::m

The number of corrections to approximate the inverse hessian matrix.

The L-BFGS routine stores the computation results of previous m iterations to approximate the inverse hessian matrix of the current iteration. This parameter controls the size of the limited memories (corrections). The default value is 6. Values less than 3 are not recommended. Large values will result in excessive computing time.

lbfgsfloatval_t lbfgs_parameter_t::epsilon

Epsilon for convergence test.

This parameter determines the accuracy with which the solution is to be found. A minimization terminates when ||g|| < epsilon * max(1, ||x||), where ||.|| denotes the Euclidean (L2) norm. The default value is 1e-5.

int lbfgs_parameter_t::max_iterations

The maximum number of iterations.

The lbfgs() function terminates an optimization process with LBFGSERR_MAXIMUMITERATION status code when the iteration count exceedes this parameter. Setting this parameter to zero continues an optimization process until a convergence or error. The default value is 0.

int lbfgs_parameter_t::linesearch

The line search algorithm.

This parameter specifies a line search algorithm to be used by the L-BFGS routine.

int lbfgs_parameter_t::max_linesearch

The maximum number of trials for the line search.

This parameter controls the number of function and gradients evaluations per iteration for the line search routine. The default value is 20.

lbfgsfloatval_t lbfgs_parameter_t::min_step

The minimum step of the line search routine.

The default value is 1e-20. This value need not be modified unless the exponents are too large for the machine being used, or unless the problem is extremely badly scaled (in which case the exponents should be increased).

lbfgsfloatval_t lbfgs_parameter_t::max_step

The maximum step of the line search.

The default value is 1e+20. This value need not be modified unless the exponents are too large for the machine being used, or unless the problem is extremely badly scaled (in which case the exponents should be increased).

lbfgsfloatval_t lbfgs_parameter_t::ftol

A parameter to control the accuracy of the line search routine.

The default value is 1e-4. This parameter should be greater than zero and smaller than 0.5.

lbfgsfloatval_t lbfgs_parameter_t::gtol

A parameter to control the accuracy of the line search routine.

The default value is 0.9. If the function and gradient evaluations are inexpensive with respect to the cost of the iteration (which is sometimes the case when solving very large problems) it may be advantageous to set this parameter to a small value. A typical small value is 0.1. This parameter shuold be greater than the ftol parameter (1e-4) and smaller than 1.0.

lbfgsfloatval_t lbfgs_parameter_t::xtol

The machine precision for floating-point values.

This parameter must be a positive value set by a client program to estimate the machine precision. The line search routine will terminate with the status code (LBFGSERR_ROUNDING_ERROR) if the relative width of the interval of uncertainty is less than this parameter.

lbfgsfloatval_t lbfgs_parameter_t::orthantwise_c

Coeefficient for the L1 norm of variables.

This parameter should be set to zero for standard minimization problems. Setting this parameter to a positive value activates Orthant-Wise Limited-memory Quasi-Newton (OWL-QN) method, which minimizes the objective function F(x) combined with the L1 norm |x| of the variables, {F(x) + C |x|}. This parameter is the coeefficient for the |x|, i.e., C. As the L1 norm |x| is not differentiable at zero, the library modifies function and gradient evaluations from a client program suitably; a client program thus have only to return the function value F(x) and gradients G(x) as usual. The default value is zero.

int lbfgs_parameter_t::orthantwise_start

Start index for computing L1 norm of the variables.

This parameter is valid only for OWL-QN method (i.e., orthantwise_c != 0). This parameter b (0 <= b < N) specifies the index number from which the library computes the L1 norm of the variables x, |x| := |x_{b}| + |x_{b+1}| + ... + |x_{N}| . In other words, variables x_1, ..., x_{b-1} are not used for computing the L1 norm. Setting b (0 < b < N), one can protect variables, x_1, ..., x_{b-1} (e.g., a bias term of logistic regression) from being regularized. The default value is zero.

int lbfgs_parameter_t::orthantwise_end

End index for computing L1 norm of the variables.

This parameter is valid only for OWL-QN method (i.e., orthantwise_c != 0). This parameter e (0 < e <= N) specifies the index number at which the library stops computing the L1 norm of the variables x,


Copyright (c) 2002-2008 by Naoaki Okazaki
Sun Nov 2 13:54:50 2008