opennn::NeuronSelection Class Reference

Abstract base class for hidden-layer-size selection methods. More...

#include <neuron_selection.h>

Inheritance diagram for opennn::NeuronSelection:

Public Types
enum class	StoppingCondition {   MaximumTime , SelectionErrorGoal , MaximumEpochs , MaximumSelectionFailures ,   MaximumNeurons }
	Reasons that can terminate a neurons-selection run. More...

Public Member Functions
	NeuronSelection (TrainingStrategy *training_strategy=nullptr)
	Constructs the selector.
virtual	~NeuronSelection ()=default
	Virtual destructor.
const TrainingStrategy *	get_training_strategy () const
	Read-only access to the bound training strategy.
bool	has_training_strategy () const
	Whether a training strategy has been bound.
bool	get_display () const
	Whether progress should be printed to stdout.
void	set (TrainingStrategy *)
	Re-initializes the selector by setting its training strategy.
void	set_training_strategy (TrainingStrategy *new_training_strategy)
	Sets the training strategy directly.
void	set_default ()
	Resets all hyperparameters to their default values.
void	set_maximum_neurons (const Index new_maximum_neurons)
	Sets the maximum hidden-layer size considered.
void	set_minimum_neurons (const Index new_minimum_neurons)
	Sets the minimum hidden-layer size considered.
void	set_trials_number (const Index new_trials_number)
	Sets the number of training trials per candidate (mean is used).
void	set_display (bool new_display)
	Toggles per-iteration progress printing.
void	set_validation_error_goal (const float new_validation_error_goal)
	Sets the validation error goal.
void	set_maximum_epochs (const Index new_maximum_epochs)
	Sets the maximum training epochs per evaluation.
void	set_maximum_validation_failures (const Index new_maximum_validation_failures)
	Sets the maximum number of consecutive validation-error increases tolerated.
void	set_maximum_time (const float new_maximum_time)
	Sets the maximum wall-clock selection time.
void	check () const
	Validates the bound configuration; throws if anything is missing.
virtual NeuronsSelectionResults	perform_neurons_selection ()=0
	Runs the selection algorithm.
string	get_name () const
	Canonical name of the selector (set by subclasses).
virtual void	from_JSON (const JsonDocument &)=0
	Loads selector hyperparameters from a parsed JSON document.
virtual void	to_JSON (JsonWriter &) const =0
	Writes selector hyperparameters to a streaming JSON writer.
void	save (const filesystem::path &) const
	Saves the selector configuration to a file.
void	load (const filesystem::path &)
	Loads the selector configuration from a file.
virtual void	print () const
	Prints a human-readable summary of the selector to stdout.

Protected Attributes
TrainingStrategy *	training_strategy = nullptr
	Training strategy used to evaluate candidates; not owned.
VectorR	validation_error_history
	Per-iteration validation error of the best candidate so far.
VectorR	training_error_history
	Per-iteration training error of the best candidate so far.
Index	minimum_neurons = 0
	Lower bound on the hidden-layer size.
Index	maximum_neurons = 0
	Upper bound on the hidden-layer size.
Index	trials_number = 1
	Number of independent training runs averaged per candidate.
float	validation_error_goal = 0
	Validation error goal; selection stops when reached.
Index	maximum_epochs = 10
	Maximum training epochs per evaluation.
Index	maximum_validation_failures = 100
	Maximum consecutive validation-error increases tolerated.
float	maximum_time = 0
	Maximum wall-clock selection time in seconds.
bool	display = true
	Whether progress should be printed to stdout during selection.
string	name
	Canonical name of the selector (set by subclasses).

Detailed Description

Abstract base class for hidden-layer-size selection methods.

Holds a pointer to a TrainingStrategy used to evaluate candidate architectures, the common stopping criteria and trial averaging. Subclasses implement perform_neurons_selection() to drive the search.

Member Enumeration Documentation

StoppingCondition

enum class opennn::NeuronSelection::StoppingCondition

strong

Reasons that can terminate a neurons-selection run.

Enumerator
MaximumTime	Configured time budget exhausted.
SelectionErrorGoal	Validation error reached the configured goal.
MaximumEpochs	Configured epoch budget exhausted.
MaximumSelectionFailures	Validation error increased too many times.
MaximumNeurons	Maximum hidden-layer size reached.

Constructor & Destructor Documentation

NeuronSelection()

opennn::NeuronSelection::NeuronSelection

(

TrainingStrategy *

training_strategy = nullptr

)

Constructs the selector.

Parameters

training_strategy

Training strategy used to evaluate candidate sizes.

~NeuronSelection()

virtual opennn::NeuronSelection::~NeuronSelection ( )

virtualdefault

Virtual destructor.

Member Function Documentation

check()

void opennn::NeuronSelection::check

(

)

const

Validates the bound configuration; throws if anything is missing.

from_JSON()

virtual void opennn::NeuronSelection::from_JSON ( const JsonDocument & )

pure virtual

Loads selector hyperparameters from a parsed JSON document.

Implemented in opennn::GrowingNeurons.

get_display()

bool opennn::NeuronSelection::get_display ( ) const

inline

Whether progress should be printed to stdout.

get_name()

string opennn::NeuronSelection::get_name ( ) const

inline

Canonical name of the selector (set by subclasses).

get_training_strategy()

const TrainingStrategy * opennn::NeuronSelection::get_training_strategy ( ) const

inline

Read-only access to the bound training strategy.

has_training_strategy()

bool opennn::NeuronSelection::has_training_strategy ( ) const

inline

Whether a training strategy has been bound.

load()

void opennn::NeuronSelection::load

(

const filesystem::path &

)

Loads the selector configuration from a file.

Receives the source path.

perform_neurons_selection()

virtual NeuronsSelectionResults opennn::NeuronSelection::perform_neurons_selection ( )

pure virtual

Runs the selection algorithm.

Returns

Best-of-run hidden-layer size and supporting statistics.

Implemented in opennn::GrowingNeurons.

print()

virtual void opennn::NeuronSelection::print ( ) const

inlinevirtual

Prints a human-readable summary of the selector to stdout.

save()

void opennn::NeuronSelection::save

(

const filesystem::path &

)

const

Saves the selector configuration to a file.

Receives the destination path.

set()

void opennn::NeuronSelection::set

(

TrainingStrategy *

)

Re-initializes the selector by setting its training strategy.

Receives the training strategy used to evaluate candidates.

set_default()

void opennn::NeuronSelection::set_default

(

)

Resets all hyperparameters to their default values.

set_display()

void opennn::NeuronSelection::set_display ( bool new_display )

inline

Toggles per-iteration progress printing.

Parameters

new_display

True to print progress to stdout.

set_maximum_epochs()

void opennn::NeuronSelection::set_maximum_epochs ( const Index new_maximum_epochs )

inline

Sets the maximum training epochs per evaluation.

Parameters

new_maximum_epochs

Epoch budget for each candidate's training.

set_maximum_neurons()

void opennn::NeuronSelection::set_maximum_neurons ( const Index new_maximum_neurons )

inline

Sets the maximum hidden-layer size considered.

Parameters

new_maximum_neurons

Upper bound on the number of neurons.

set_maximum_time()

void opennn::NeuronSelection::set_maximum_time ( const float new_maximum_time )

inline

Sets the maximum wall-clock selection time.

Parameters

new_maximum_time

Time budget in seconds.

set_maximum_validation_failures()

void opennn::NeuronSelection::set_maximum_validation_failures ( const Index new_maximum_validation_failures )

inline

Sets the maximum number of consecutive validation-error increases tolerated.

Parameters

new_maximum_validation_failures

Failure budget for early stopping.

set_minimum_neurons()

void opennn::NeuronSelection::set_minimum_neurons ( const Index new_minimum_neurons )

inline

Sets the minimum hidden-layer size considered.

Parameters

new_minimum_neurons

Lower bound on the number of neurons.

set_training_strategy()

void opennn::NeuronSelection::set_training_strategy ( TrainingStrategy * new_training_strategy )

inline

Sets the training strategy directly.

Parameters

new_training_strategy

Strategy used to evaluate candidates.

set_trials_number()

void opennn::NeuronSelection::set_trials_number ( const Index new_trials_number )

inline

Sets the number of training trials per candidate (mean is used).

Parameters

new_trials_number

Number of independent training runs averaged.

set_validation_error_goal()

void opennn::NeuronSelection::set_validation_error_goal ( const float new_validation_error_goal )

inline

Sets the validation error goal.

Parameters

new_validation_error_goal

Selection stops when validation error reaches this value.

to_JSON()

virtual void opennn::NeuronSelection::to_JSON ( JsonWriter & ) const

pure virtual

Writes selector hyperparameters to a streaming JSON writer.

Implemented in opennn::GrowingNeurons.

Member Data Documentation

display

bool opennn::NeuronSelection::display = true

protected

Whether progress should be printed to stdout during selection.

maximum_epochs

Index opennn::NeuronSelection::maximum_epochs = 10

protected

Maximum training epochs per evaluation.

maximum_neurons

Index opennn::NeuronSelection::maximum_neurons = 0

protected

Upper bound on the hidden-layer size.

maximum_time

float opennn::NeuronSelection::maximum_time = 0

protected

Maximum wall-clock selection time in seconds.

maximum_validation_failures

Index opennn::NeuronSelection::maximum_validation_failures = 100

protected

Maximum consecutive validation-error increases tolerated.

minimum_neurons

Index opennn::NeuronSelection::minimum_neurons = 0

protected

Lower bound on the hidden-layer size.

name

string opennn::NeuronSelection::name

protected

Canonical name of the selector (set by subclasses).

training_error_history

VectorR opennn::NeuronSelection::training_error_history

protected

Per-iteration training error of the best candidate so far.

training_strategy

TrainingStrategy* opennn::NeuronSelection::training_strategy = nullptr

protected

Training strategy used to evaluate candidates; not owned.

trials_number

Index opennn::NeuronSelection::trials_number = 1

protected

Number of independent training runs averaged per candidate.

validation_error_goal

float opennn::NeuronSelection::validation_error_goal = 0

protected

Validation error goal; selection stops when reached.

validation_error_history

VectorR opennn::NeuronSelection::validation_error_history

protected

Per-iteration validation error of the best candidate so far.