Abstract base class for vector-valued functions \(g\colon [0, 1]^d \to \mathbb{R}^m\) together with their Jacobians \(\nabla g\colon [0, 1]^d \to \mathbb{R}^{m \times d}\), i.e. More...

#include <VectorFunctionHessian.hpp>

Inheritance diagram for sgpp::optimization::VectorFunctionHessian:

Public Member Functions
virtual void	clone (std::unique_ptr< VectorFunctionHessian > &clone) const =0
	Pure virtual method for cloning the Hessian. More...

virtual void	eval (const base::DataVector &x, base::DataVector &value, base::DataMatrix &gradient, std::vector< base::DataMatrix > &hessian)=0
	Pure virtual method for calculating \(g(\vec{x})\) together with \(\nabla g(\vec{x})\). More...

size_t	getNumberOfComponents () const

size_t	getNumberOfParameters () const

	VectorFunctionHessian (size_t d, size_t m)
	Constructor. More...

virtual	~VectorFunctionHessian ()
	Destructor. More...

Protected Attributes
size_t	d
	dimension of the domain More...

size_t	m
	number of components More...

Detailed Description

Abstract base class for vector-valued functions \(g\colon [0, 1]^d \to \mathbb{R}^m\) together with their Jacobians \(\nabla g\colon [0, 1]^d \to \mathbb{R}^{m \times d}\), i.e.

\((\nabla g)_{i,t} = \frac{\partial}{\partial x_t} g_i\) with \(g = (g_i)_{i=1}^m\), and their Hessians \(\nabla g\colon [0, 1]^d \to \mathbb{R}^{m \times d \times d}\), i.e. \((\nabla^2 g)_{i,t_1,t_2} = \frac{\partial^2}{\partial x_{t_1} x_{t_2}} g_i\) (e.g., equality/inequality constraints \(g(\vec{x}) \le \vec{0}\) or \(g(\vec{x}) = \vec{0}\) in optimization).

Constructor & Destructor Documentation

◆ VectorFunctionHessian()

sgpp::optimization::VectorFunctionHessian::VectorFunctionHessian	(	size_t	d,
		size_t	m
	)

inline

Constructor.

Parameters

d	dimension of the domain
m	number of components

◆ ~VectorFunctionHessian()

virtual sgpp::optimization::VectorFunctionHessian::~VectorFunctionHessian ( )

inlinevirtual

Destructor.

References eval().

Member Function Documentation

◆ clone()

virtual void sgpp::optimization::VectorFunctionHessian::clone ( std::unique_ptr< VectorFunctionHessian > & clone ) const

pure virtual

Pure virtual method for cloning the Hessian.

It should generate a pointer to the cloned object and it's used for parallel computations (the eval() method might not be thread-safe).

Parameters

[out] clone pointer to cloned object

Implemented in sgpp::optimization::InterpolantVectorFunctionHessian, and sgpp::optimization::WrapperVectorFunctionHessian.

Referenced by getNumberOfComponents().

◆ eval()

virtual void sgpp::optimization::VectorFunctionHessian::eval	(	const base::DataVector &	x,
		base::DataVector &	value,
		base::DataMatrix &	gradient,
		std::vector< base::DataMatrix > &	hessian
	)

pure virtual

Pure virtual method for calculating \(g(\vec{x})\) together with \(\nabla g(\vec{x})\).

Parameters

[in]	x	evaluation point \(\vec{x} \in [0, 1]^d\)
[out]	value	\(g(\vec{x})\)
[out]	gradient	Jacobian \(\nabla g(\vec{x}) \in \mathbb{R}^{m \times d}\)
[out]	hessian	\(m\)-vector of Hessians \(\nabla^2 g_i(\vec{x}) \in \mathbb{R}^{d \times d}\)