All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Modules Pages
NnetCombineAconfig Struct Reference

#include <combine-nnet-a.h>

Collaboration diagram for NnetCombineAconfig:

Public Member Functions

 NnetCombineAconfig ()
 
void Register (OptionsItf *opts)
 

Public Attributes

int32 num_bfgs_iters
 
BaseFloat initial_step
 
BaseFloat valid_impr_thresh
 
BaseFloat overshoot
 
BaseFloat min_learning_rate_factor
 
BaseFloat max_learning_rate_factor
 
BaseFloat min_learning_rate
 

Detailed Description

Definition at line 31 of file combine-nnet-a.h.

Constructor & Destructor Documentation

Member Function Documentation

void Register ( OptionsItf opts)
inline

Definition at line 52 of file combine-nnet-a.h.

References OptionsItf::Register().

Referenced by main().

52  {
53  opts->Register("num-bfgs-iters", &num_bfgs_iters, "Maximum number of function "
54  "evaluations for BFGS to use when optimizing combination weights");
55  opts->Register("initial-step", &initial_step, "Parameter in the optimization, "
56  "used to set the initial step length; the default value should be "
57  "suitable.");
58  opts->Register("num-bfgs-iters", &num_bfgs_iters, "Maximum number of function "
59  "evaluations for BFGS to use when optimizing combination weights");
60  opts->Register("valid-impr-thresh", &valid_impr_thresh, "Threshold of improvement "
61  "in validation-set objective function for one iteratin; below this, "
62  "we start using the \"overshoot\" mechanism to keep learning rates high.");
63  opts->Register("overshoot", &overshoot, "Factor by which we overshoot the step "
64  "size obtained by BFGS; only applies when validation set impr is less "
65  "than valid-impr-thresh.");
66  opts->Register("max-learning-rate-factor", &max_learning_rate_factor,
67  "Maximum factor by which to increase the learning rate for any layer.");
68  opts->Register("min-learning-rate-factor", &min_learning_rate_factor,
69  "Minimum factor by which to increase the learning rate for any layer.");
70  opts->Register("min-learning-rate", &min_learning_rate,
71  "Floor on the automatically updated learning rates");
72  }

Member Data Documentation

BaseFloat initial_step

Definition at line 36 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

BaseFloat max_learning_rate_factor

Definition at line 42 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

BaseFloat min_learning_rate

Definition at line 43 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

BaseFloat min_learning_rate_factor

Definition at line 41 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

int32 num_bfgs_iters

Definition at line 32 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

BaseFloat overshoot

Definition at line 39 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().

BaseFloat valid_impr_thresh

Definition at line 38 of file combine-nnet-a.h.

Referenced by kaldi::nnet2::CombineNnetsA().


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