Optimizes input values so that a network's outputs satisfy user-defined conditions and objectives. More...

#include <response_optimization.h>

Classes
struct	Condition
	Constraint or objective imposed on a single variable, with optional bounds. More...

struct	Domain
	Bounded domain in input or output space defined by inferior and superior frontiers. More...

struct	Objectives
	Encodes the objectives extracted from the response optimization configuration. More...

Public Types
enum class	ConditionType { None , Between , EqualTo , LessEqualTo , GreaterEqualTo , LessThan , GreaterThan , Minimize , Maximize }
	Types of constraint or objective applied to an input or output variable. More...

Public Member Functions
Objectives	build_objectives () const
	Builds the Objectives helper from the current conditions.

	ResponseOptimization (NeuralNetwork =nullptr, Dataset =nullptr)
	Constructs the optimizer bound to an optional neural network and dataset.

void	set (NeuralNetwork =nullptr, Dataset =nullptr)
	Binds the optimizer to a neural network and a dataset.

void	clear_conditions ()
	Removes all conditions previously configured on input and output variables.

void	set_condition (const string &name, const ConditionType condition, float low_bound=0.0, float up_bound=0.0)
	Adds or replaces the condition applied to the variable with the given name.

void	set_iterations (const int iterations)

void	set_zoom_factor (float new_zoom_factor)

void	set_evaluations_number (const int new_evaluations_number)

void	set_relative_tolerance (float new_relative_tolerance)

vector< float >	get_utopian_point () const
	Returns the coordinates of the utopian point used as reference for multiobjective optimization.

Domain	get_original_domain (const string role) const
	Returns the original (untrimmed) domain for variables playing the given role.

Condition	get_condition (const Index index) const
	Returns the configured condition at the given index.

MatrixR	calculate_random_inputs (const Domain &input_domain) const
	Draws a random sample of input points within the given input domain.

pair< MatrixR, MatrixR >	filter_feasible_points (const MatrixR &inputs, const MatrixR &outputs, const Domain &output_domain) const
	Filters the (inputs, outputs) pairs whose outputs fall inside the feasible output domain.

pair< MatrixR, MatrixR >	calculate_optimal_points (const MatrixR &feasible_inputs, const MatrixR &feasible_outputs, const Objectives &objectives) const
	Selects the optimal points among feasible candidates according to the supplied objectives.

MatrixR	assemble_results (const MatrixR &inputs, const MatrixR &outputs) const
	Assembles the final results matrix concatenating inputs and outputs.

MatrixR	perform_single_objective_optimization (const Objectives &objectives) const
	Performs single-objective optimization and returns the best input/output pair.

pair< MatrixR, MatrixR >	calculate_pareto (const MatrixR &inputs, const MatrixR &outputs, const MatrixR &objective_matrix) const
	Computes the Pareto front of the supplied input/output samples.

pair< float, float >	calculate_quality_metrics (const MatrixR &inputs, const MatrixR &outputs, const Objectives &objectives) const
	Computes quality metrics of the optimization (e.g. distance to utopian point).

MatrixR	perform_multiobjective_optimization (const Objectives &objectives) const
	Performs multiobjective optimization and returns the Pareto-optimal input/output samples.

MatrixR	perform_response_optimization () const
	Runs the response optimization using the configured conditions and returns the optimal results.

Detailed Description

Optimizes input values so that a network's outputs satisfy user-defined conditions and objectives.

Member Enumeration Documentation

◆ ConditionType

enum class opennn::ResponseOptimization::ConditionType

strong

Types of constraint or objective applied to an input or output variable.

Enumerator
None
Between
EqualTo
LessEqualTo
GreaterEqualTo
LessThan
GreaterThan
Minimize
Maximize

Constructor & Destructor Documentation

◆ ResponseOptimization()

opennn::ResponseOptimization::ResponseOptimization	(	NeuralNetwork *	= nullptr,
		Dataset *	= nullptr )

Constructs the optimizer bound to an optional neural network and dataset.

Member Function Documentation

◆ assemble_results()

MatrixR opennn::ResponseOptimization::assemble_results	(	const MatrixR &	inputs,
		const MatrixR &	outputs ) const

Assembles the final results matrix concatenating inputs and outputs.

◆ build_objectives()

Objectives opennn::ResponseOptimization::build_objectives ( ) const

Builds the Objectives helper from the current conditions.

◆ calculate_optimal_points()

pair< MatrixR, MatrixR > opennn::ResponseOptimization::calculate_optimal_points	(	const MatrixR &	feasible_inputs,
		const MatrixR &	feasible_outputs,
		const Objectives &	objectives ) const

Selects the optimal points among feasible candidates according to the supplied objectives.

◆ calculate_pareto()

pair< MatrixR, MatrixR > opennn::ResponseOptimization::calculate_pareto	(	const MatrixR &	inputs,
		const MatrixR &	outputs,
		const MatrixR &	objective_matrix ) const

Computes the Pareto front of the supplied input/output samples.

◆ calculate_quality_metrics()

pair< float, float > opennn::ResponseOptimization::calculate_quality_metrics	(	const MatrixR &	inputs,
		const MatrixR &	outputs,
		const Objectives &	objectives ) const

Computes quality metrics of the optimization (e.g. distance to utopian point).

◆ calculate_random_inputs()

MatrixR opennn::ResponseOptimization::calculate_random_inputs ( const Domain & input_domain ) const

Draws a random sample of input points within the given input domain.

◆ clear_conditions()

void opennn::ResponseOptimization::clear_conditions ( )

Removes all conditions previously configured on input and output variables.

◆ filter_feasible_points()

pair< MatrixR, MatrixR > opennn::ResponseOptimization::filter_feasible_points	(	const MatrixR &	inputs,
		const MatrixR &	outputs,
		const Domain &	output_domain ) const

Filters the (inputs, outputs) pairs whose outputs fall inside the feasible output domain.

◆ get_condition()

Condition opennn::ResponseOptimization::get_condition ( const Index index ) const

Returns the configured condition at the given index.

◆ get_original_domain()

Domain opennn::ResponseOptimization::get_original_domain ( const string role ) const

Returns the original (untrimmed) domain for variables playing the given role.

◆ get_utopian_point()

vector< float > opennn::ResponseOptimization::get_utopian_point ( ) const

Returns the coordinates of the utopian point used as reference for multiobjective optimization.

◆ perform_multiobjective_optimization()

MatrixR opennn::ResponseOptimization::perform_multiobjective_optimization ( const Objectives & objectives ) const

Performs multiobjective optimization and returns the Pareto-optimal input/output samples.

◆ perform_response_optimization()

MatrixR opennn::ResponseOptimization::perform_response_optimization ( ) const

Runs the response optimization using the configured conditions and returns the optimal results.

◆ perform_single_objective_optimization()

MatrixR opennn::ResponseOptimization::perform_single_objective_optimization ( const Objectives & objectives ) const

Performs single-objective optimization and returns the best input/output pair.

◆ set()

void opennn::ResponseOptimization::set	(	NeuralNetwork *	= nullptr,
		Dataset *	= nullptr )

Binds the optimizer to a neural network and a dataset.

◆ set_condition()

void opennn::ResponseOptimization::set_condition	(	const string &	name,
		const ConditionType	condition,
		float	low_bound = 0.0,
		float	up_bound = 0.0 )

Adds or replaces the condition applied to the variable with the given name.

Parameters

name	Variable name as defined in the dataset.
condition	Condition type to enforce.
low_bound	Lower bound used by bounded condition types.
up_bound	Upper bound used by bounded condition types.

◆ set_evaluations_number()

void opennn::ResponseOptimization::set_evaluations_number ( const int new_evaluations_number )

◆ set_iterations()

void opennn::ResponseOptimization::set_iterations ( const int iterations )

◆ set_relative_tolerance()

void opennn::ResponseOptimization::set_relative_tolerance ( float new_relative_tolerance )

◆ set_zoom_factor()

void opennn::ResponseOptimization::set_zoom_factor ( float new_zoom_factor )

Classes

Public Types

Public Member Functions