lbann::objective_function Class Reference

#include <objective_function.hpp>

Public Member Functions
	objective_function ()
	objective_function (const objective_function &other)
objective_function &	operator= (const objective_function &other)
	~objective_function ()=default
objective_function *	copy () const
template<class Archive >
void	serialize (Archive &ar)
void	add_term (std::unique_ptr< objective_function_term > term)
std::vector< objective_function_term * >	get_terms ()
void	setup (model &m)
void	start_evaluation (execution_mode mode, int mini_batch_size)
EvalType	finish_evaluation (execution_mode mode, int mini_batch_size)
void	differentiate ()
void	compute_weight_regularization ()
void	reset_statistics ()
void	reset_statistics (execution_mode mode)
EvalType	get_mean_value (execution_mode mode) const
int	get_statistics_num_samples (execution_mode mode) const
std::vector< ViewingLayerPtr >	get_layer_pointers () const
void	set_layer_pointers (std::vector< ViewingLayerPtr > layers)
std::vector< ViewingWeightsPtr >	get_weights_pointers () const
void	set_weights_pointers (std::vector< ViewingWeightsPtr > w)
EvalType	get_evaluation_time () const
EvalType	get_differentiation_time () const
void	reset_counters ()
void	write_proto (lbann_data::ObjectiveFunction &proto) const

Private Attributes
std::vector< std::unique_ptr< objective_function_term > >	m_terms
std::map< execution_mode, metric_statistics >	m_statistics
EvalType	m_evaluation_time = EvalType(0)
EvalType	m_differentiation_time = EvalType(0)

Detailed Description

Objective function class.

Definition at line 36 of file objective_function.hpp.

Constructor & Destructor Documentation

objective_function() [1/2]

lbann::objective_function::objective_function ( )

inline

Default constructor.

Definition at line 40 of file objective_function.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

objective_function() [2/2]

lbann::objective_function::objective_function

(

const objective_function &

other

)

Copy constructor.

~objective_function()

lbann::objective_function::~objective_function ( )

default

Destructor.

Here is the caller graph for this function:

Member Function Documentation

add_term()

void lbann::objective_function::add_term

(

std::unique_ptr< objective_function_term >

term

)

Add a term to the objective function.

Here is the caller graph for this function:

compute_weight_regularization()

void lbann::objective_function::compute_weight_regularization

(

)

Compute the gradient of the weight regularization term. The gradient is computed w.r.t. the weights.

Here is the caller graph for this function:

copy()

objective_function* lbann::objective_function::copy ( ) const

inline

Copy function.

Definition at line 49 of file objective_function.hpp.

Here is the call graph for this function:

differentiate()

void lbann::objective_function::differentiate

(

)

Compute the objective function gradient. The gradient is with respect to the objective function inputs

Here is the caller graph for this function:

finish_evaluation()

EvalType lbann::objective_function::finish_evaluation	(	execution_mode	mode,
		int	mini_batch_size
	)

Complete evaluation of the objective function. The result is stored in history.

Here is the caller graph for this function:

get_differentiation_time()

EvalType lbann::objective_function::get_differentiation_time ( ) const

inline

Get the time spent computing the objective function gradient.

Definition at line 117 of file objective_function.hpp.

get_evaluation_time()

EvalType lbann::objective_function::get_evaluation_time ( ) const

inline

Get the time spent evaluating the objective function.

Definition at line 115 of file objective_function.hpp.

get_layer_pointers()

std::vector<ViewingLayerPtr> lbann::objective_function::get_layer_pointers

(

)

const

Get list of pointers to layers.

Here is the caller graph for this function:

get_mean_value()

EvalType lbann::objective_function::get_mean_value

(

execution_mode

mode

)

const

Get mean objective function value. This is a weighted average such that each mini-batch sample makes an equal contribution.

Here is the caller graph for this function:

get_statistics_num_samples()

int lbann::objective_function::get_statistics_num_samples

(

execution_mode

mode

)

const

Get number of samples for statistics.

Here is the caller graph for this function:

get_terms()

std::vector<objective_function_term*> lbann::objective_function::get_terms

(

)

Get list of objective function terms.

Here is the caller graph for this function:

get_weights_pointers()

std::vector<ViewingWeightsPtr> lbann::objective_function::get_weights_pointers

(

)

const

Get list of pointers to weights.

Here is the caller graph for this function:

operator=()

objective_function& lbann::objective_function::operator=

(

const objective_function &

other

)

Copy assignment operator.

Here is the caller graph for this function:

reset_counters()

void lbann::objective_function::reset_counters ( )

inline

Reset time counters.

Definition at line 119 of file objective_function.hpp.

Here is the call graph for this function:

reset_statistics() [1/2]

void lbann::objective_function::reset_statistics ( )

inline

Clear all statistics.

Definition at line 88 of file objective_function.hpp.

reset_statistics() [2/2]

void lbann::objective_function::reset_statistics ( execution_mode  mode )

inline

Clear statistics for an execution mode.

Definition at line 95 of file objective_function.hpp.

Here is the call graph for this function:

serialize()

template<class Archive >

void lbann::objective_function::serialize

(

Archive &

ar

)

Archive for checkpoint and restart

Here is the caller graph for this function:

set_layer_pointers()

void lbann::objective_function::set_layer_pointers

(

std::vector< ViewingLayerPtr >

layers

)

Set list of pointers to layers.

Here is the caller graph for this function:

set_weights_pointers()

void lbann::objective_function::set_weights_pointers

(

std::vector< ViewingWeightsPtr >

w

)

Set list of pointers to weights.

Here is the caller graph for this function:

setup()

void lbann::objective_function::setup

(

model &

m

)

Setup objective function.

Here is the caller graph for this function:

start_evaluation()

void lbann::objective_function::start_evaluation	(	execution_mode	mode,
		int	mini_batch_size
	)

Start evaluating the objective function. This function takes the model's current mini-batch size. If multiple models are being trained, the current mini-batch size may be different from the effective mini-batch size. The result is not guaranteed to be available until finish_evaluation is called.

Here is the caller graph for this function:

write_proto()

void lbann::objective_function::write_proto

(

lbann_data::ObjectiveFunction &

proto

)

const

Add Objection Function data to prototext

Here is the caller graph for this function:

Member Data Documentation

m_differentiation_time

EvalType lbann::objective_function::m_differentiation_time = EvalType(0)

private

Time spent computing the objective function gradient.

Definition at line 138 of file objective_function.hpp.

m_evaluation_time

EvalType lbann::objective_function::m_evaluation_time = EvalType(0)

private

Time spent evaluating the objective function.

Definition at line 136 of file objective_function.hpp.

m_statistics

std::map<execution_mode, metric_statistics> lbann::objective_function::m_statistics

private

Objective funciton statistics.

Definition at line 133 of file objective_function.hpp.

m_terms

std::vector<std::unique_ptr<objective_function_term> > lbann::objective_function::m_terms

private

List of objective function terms.

Definition at line 130 of file objective_function.hpp.

---

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

• objective_function.hpp