DualVector< T, N > Class Template Reference

#include <vigra/autodiff.hxx>

Public Types
typedef TinyVector< T, N >	Gradient
	type of the gradient vector
typedef T	value_type
	type of function values and gradient elements

Public Member Functions
	DualVector ()
	DualVector (T const &val)
	DualVector (T const &val, Gradient const &grad)
	DualVector (T const &val, T const &g0)
	DualVector (T const &val, T const &g0, T const &g1)
	DualVector (T const &val, int targetElement)
Gradient const &	gradient () const
T	value () const

Detailed Description

template<class T, int N>
class vigra::autodiff::DualVector< T, N >

Number type for automatic differentiation.

Automatic differentiation allows one to compute the value of a numeric expression and its gradient with respect to the expression's arguments automatically and in one go. To support this, one needs a special number type that holds a scalar value and the corresponding gradient vector of appropriate length. This is the purpose of hte template class DualVector<T, N>, where T is the underlying numerical type (usually 'double'), and N denotes the length of the gradient vector.

The standard arithmetic and algebraic functions are overloaded for DualVector in order to implement the required arithmetic of dual numbers. When you replace all arguments in a numeric expression with the appropriate DualVector instances, the result will be a DualVector that contains the result value and gradient of the expression, evaluated at the point defined by the input values.

Usage:

#include <vigra/autodiff.hxx>
Namespace: vigra::autodiff

typedef DualVector<double, 2> N;  // for expressions with two arguments

N x(1.0, 0);  // first argument of the expression
N s(2.0, 1);  // second argument of the expression

N y = exp(-0.5 * sq(x / s));

std::cout << "Evaluated exp(- x^2 / (2 s^2)) at x=1 and s = 2:\n";
std::cout << "result = " << y.value() <<", gradient = " << y.gradient() << "\n";

Note that the second argument of the DualVector constructors specifies that the derivative w.r.t 'x' shall be the element 0 of the gradient vector, and the derivative w.r.t. 's' shall be element 1.

Constructor & Destructor Documentation

DualVector

(

)

Zero initialization.

DualVector ( T const &  val )

explicit

Provide a value, but zero-initialize the gradient.

DualVector	(	T const &	val,
		Gradient const &	grad
	)

Initialize with given value and gradient.

DualVector	(	T const &	val,
		T const &	g0
	)

Shorthand for DualVector(val, Gradient(g0)) when N == 1.

Not to be used when N != 1.

DualVector	(	T const &	val,
		T const &	g0,
		T const &	g1
	)

Shorthand for DualVector(val, Gradient(g0, g1)) when N == 2.

Not to be used when N != 2.

DualVector	(	T const &	val,
		int	targetElement
	)

Initialize value to represent the argument number 'targetElement' in an expression.

The derivative of the expression w.r.t. this variable will be element 'targetElement' of the resulting gradient vector.

Member Function Documentation

T value

(

)

const

Get current value.

Gradient const& gradient

(

)

const

Get current gradient.

---

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

• vigra/autodiff.hxx