cpp_gradient¶

C++ Abstract Class For Calculating Gradient¶

Syntax¶

      # include <cmpad/gradient.hpp>
      template< template<class ADVector > class Algo > class
      Grad : public cmpad::gradient {
            …
      };

Source Code¶# include <cmpad/vector.hpp>
# include <cmpad/fun_obj.hpp>

namespace cmpad {
   // gradient
   class gradient : public fun_obj< cmpad::vector<double> > {
   public:
      // scalar_type
      typedef double scalar_type;
      // setup
      virtual void setup(const option_t& option) override = 0;
      // option
      virtual const option_t& option(void) const override = 0;
      // domain
      virtual size_t domain(void) const override  = 0;
      // range
      size_t range(void) const override
      {  return domain(); }
      // operator
      virtual const cmpad::vector<double>& operator()(
         const cmpad::vector<double>& x
      ) override = 0;
   };

}

ADVector¶

This is a Vector type chosen by the package. The type ADVector::value_type is the scalar type used by the package to record the operations in the algorithm.

Algo¶

The Algo class is derived from the cpp_fun_obj class (we use algo for a corresponding object). The gradient is for the last component of the range space of the algorithm. Hence, algo.range() can be greater than one.

setup¶

An Algo object is initialized using its setup member function.

Grad¶

This is a cpp_fun_obj interface to the gradient of the function corresponding to Algo (we use grad for a corresponding object). The object grad calculates the gradient of the last component of the vector returned by algo.

setup¶

The object grad is initialized using its setup member function (which in turn initializes algo using its setup member function). The setup functions should do calculations that do not depend on x (to make the evaluation of the gradient faster).

vector_type¶

The type of the vectors x and g is

typedef cmpad::vector<double> Grad :: vector_type

This is different from Algo :: vector_type which usually is an AD vector type for a particular package..

scalar_type¶

The type of the elements of x and g is

typedef double Grad :: scalar_type

This is different from Algo :: scalar_type which usually is an AD scalar type for a particular package.

domain¶

This returns the dimension of the domain space dimension for the function and gradient.

range¶

This returns the dimension of the range space dimension for the gradient. The dimension of the range space for the gradient is equal to the dimension of the domain space for the gradient and function. Hence range is implemented by this interface and not virtual.

x¶

This vector has size equal to the dimension of the domain space. It is the point at which the gradient, of the function represented by algo , is evaluated.

g¶

This result is has equal to the dimension of the range space for the gradient, which is equal to the size of x . It is the gradient evaluated at the point x .

Example and Derived Classes¶

Name	Title
xam_gradient	Example and Test of Gradient
adept_gradient.hpp	Calculate Gradient Using Adept
adolc_gradient.hpp	Calculate Gradient Using ADOL-C
autodiff_gradient.hpp	Calculate Gradient Using autodiff
codi_gradient.hpp	Calculate Gradient Using CoDiPack
cppad_gradient	Gradients Using CppAD
cppad_jit_gradient.hpp	Calculate Gradient Using CppAD Jit
cppadcg_gradient.hpp	Calculate Gradient Using CppAD CodeGen
sacado_gradient.hpp	Calculate Gradient Using Sacado
xad_gradient.hpp	Calculate Gradient Using XAD