LbfgsOptions Struct Reference

This is an implementation of L-BFGS. More...

#include <optimization.h>

Collaboration diagram for LbfgsOptions:

Public Member Functions
	LbfgsOptions (bool minimize=true)

Public Attributes
bool	minimize
int	m
float	first_step_learning_rate
float	first_step_length
float	first_step_impr
float	c1
float	c2
float	d
int	max_line_search_iters
int	avg_step_length

Detailed Description

This is an implementation of L-BFGS.

It pushes responsibility for determining when to stop, onto the user. There is no call-back here: everything is done via calls to the class itself (see the example in matrix-lib-test.cc). This does not implement constrained L-BFGS, but it will handle constrained problems correctly as long as the function approaches +infinity (or -infinity for maximization problems) when it gets close to the bound of the constraint. In these types of problems, you just let the function value be +infinity for minimization problems, or -infinity for maximization problems, outside these bounds).

Definition at line 84 of file optimization.h.

Constructor & Destructor Documentation

LbfgsOptions()

LbfgsOptions ( bool  minimize = true )

inline

Definition at line 107 of file optimization.h.

                                     :
      minimize(minimize),
      m(10),
      first_step_learning_rate(1.0),
      first_step_length(0.0),
      first_step_impr(0.0),
      c1(1.0e-04),
      c2(0.9),
      d(2.0),
      max_line_search_iters(50),
      avg_step_length(4) { }

Member Data Documentation

avg_step_length

int avg_step_length

Definition at line 104 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::RecordStepLength().

c1

float c1

Definition at line 99 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::StepSizeIteration().

c2

float c2

Definition at line 100 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::StepSizeIteration().

d

float d

Definition at line 101 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::ComputeNewDirection().

first_step_impr

float first_step_impr

Definition at line 94 of file optimization.h.

Referenced by kaldi::nnet2::CombineNnets(), OptimizeLbfgs< Real >::ComputeHifNeeded(), and FastNnetCombiner::FastNnetCombiner().

first_step_learning_rate

float first_step_learning_rate

Definition at line 87 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::ComputeHifNeeded().

first_step_length

float first_step_length

Definition at line 90 of file optimization.h.

Referenced by kaldi::nnet2::CombineNnetsA(), OptimizeLbfgs< Real >::ComputeHifNeeded(), and kaldi::nnet2::ShrinkNnet().

m

int m

Definition at line 86 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::AcceptStep(), kaldi::nnet2::CombineNnets(), kaldi::nnet2::CombineNnetsA(), FastNnetCombiner::FastNnetCombiner(), OptimizeLbfgs< Real >::OptimizeLbfgs(), and kaldi::nnet2::ShrinkNnet().

max_line_search_iters

int max_line_search_iters

Definition at line 103 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::StepSizeIteration().

minimize

bool minimize

Definition at line 85 of file optimization.h.

Referenced by OptimizeLbfgs< Real >::AcceptStep(), kaldi::nnet2::CombineNnets(), kaldi::nnet2::CombineNnetsA(), OptimizeLbfgs< Real >::ComputeHifNeeded(), OptimizeLbfgs< Real >::ComputeNewDirection(), OptimizeLbfgs< Real >::DoStep(), FastNnetCombiner::FastNnetCombiner(), OptimizeLbfgs< Real >::OptimizeLbfgs(), kaldi::nnet2::ShrinkNnet(), OptimizeLbfgs< Real >::StepSizeIteration(), LogisticRegression::TrainParameters(), and kaldi::UnitTestLbfgs().

---

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

• matrix/optimization.h