#include <rescale-nnet.h>

Collaboration diagram for NnetRescaleConfig:

Public Member Functions
	NnetRescaleConfig ()

void	Register (OptionsItf *opts)

Public Attributes
BaseFloat	target_avg_deriv

BaseFloat	target_first_layer_avg_deriv

BaseFloat	target_last_layer_avg_deriv

BaseFloat	num_iters

BaseFloat	delta

BaseFloat	max_change

BaseFloat	min_change

Detailed Description

Definition at line 36 of file rescale-nnet.h.

Constructor & Destructor Documentation

◆ NnetRescaleConfig()

NnetRescaleConfig ( )

inline

Definition at line 50 of file rescale-nnet.h.

                      : target_avg_deriv(0.2),
                        target_first_layer_avg_deriv(0.3),
                        target_last_layer_avg_deriv(0.1),
                        num_iters(10),
                        delta(0.01),
                        max_change(0.2), min_change(1.0e-05) { }

Member Function Documentation

◆ Register()

void Register ( OptionsItf * opts )

inline

Definition at line 57 of file rescale-nnet.h.

References OptionsItf::Register(), and kaldi::nnet2::RescaleNnet().

                                   {
     opts->Register("target-avg-deriv", &target_avg_deriv, "Target average derivative "
                    "for hidden layers that are the not the first or last hidden layer "
                    "(as fraction of maximum derivative of the nonlinearity)");
     opts->Register("target-first-layer-avg-deriv", &target_first_layer_avg_deriv,
                    "Target average derivative for the first hidden layer"
                    "(as fraction of maximum derivative of the nonlinearity)");
     opts->Register("target-last-layer-avg-deriv", &target_last_layer_avg_deriv,
                    "Target average derivative for the last hidden layer, if "
                    "#hid-layers > 1"
                    "(as fraction of maximum derivative of the nonlinearity)");
   }  