Auto-Derive function wrapper class template. More...
#include <auto_derive.hpp>
Classes | |
| class | Evaluator |
Public Types | |
| typedef Function_::ImageType | ImageType |
| specify our image type More... | |
Public Member Functions | |
| template<typename... Args_> | |
| AutoDerive (Args_ &&... args) | |
| Standard constructor. More... | |
| void | config_grad_extrapol (DataType_ initial_h, int max_steps) |
| Configures the Gradient extrapolation scheme. More... | |
| void | config_hess_extrapol (DataType_ initial_h, int max_steps) |
| Configures the Hessian extrapolation scheme. More... | |
Static Public Attributes | |
| static constexpr bool | can_grad = true |
| we provide function gradients More... | |
| static constexpr bool | can_hess = true |
| we provide function hessiants More... | |
| static constexpr bool | can_value = true |
| our base class provides function values More... | |
| static constexpr int | domain_dim = Function_::domain_dim |
| the input function must support value computation More... | |
Protected Attributes | |
| DataType_ | _init_grad_h |
| initial h for gradient extrapolation More... | |
| DataType_ | _init_hess_h |
| initial h for hessian extrapolation More... | |
| int | _max_grad_steps |
| maximum number of gradient extrapolation steps More... | |
| int | _max_hess_steps |
| maximum number of hessian extrapolation steps More... | |
Detailed Description
class FEAT::Analytic::AutoDerive< Function_, DataType_ >
Auto-Derive function wrapper class template.
This class extends another given function by adding the capability of computing gradients and hessians via Richardson extrapolation applied onto second-order central difference quotients. The initial 'h' for the difference quotient as well as the maximum number of extrapolation steps can be adjusted by using the config_grad_extrapol() and config_hess_extrapol() functions, respectively.
- Note
- If the original function already supports the computation of gradients and/or hessians, then the original function's computation is used instead of the extrapolation scheme. Example: If the input function supports computation of function values and gradients but not hessians, then this class template additionally provides the numerical computation of hessians, but uses the original function's evaluator for the computation of gradients.
- Template Parameters
-
Function_ The function to which the numerical computation of gradients and/or hessians is to be added. This function will be used as a base-class for this template instance. DataType_ The floating point data type to be used to store the values of the initial h.
Definition at line 90 of file auto_derive.hpp.
Member Typedef Documentation
◆ ImageType
| typedef Function_::ImageType FEAT::Analytic::AutoDerive< Function_, DataType_ >::ImageType |
specify our image type
Definition at line 101 of file auto_derive.hpp.
Constructor & Destructor Documentation
◆ AutoDerive()
|
inlineexplicit |
Standard constructor.
This constructor configures the Richardson extrapolation steps for both the gradient and hessian evaluation to:
- initial h = 1E-2
- maximum steps = 10 which should be sufficient in most (non-exotic) cases.
- Parameters
-
[in] args The set of arguments to be forwarded to the original function's constructor.
Definition at line 483 of file auto_derive.hpp.
Member Function Documentation
◆ config_grad_extrapol()
|
inline |
Configures the Gradient extrapolation scheme.
This function configures the Richardson extrapolation scheme for the evaluation of function gradients (first-order derivatives).
- Parameters
-
[in] initial_h The initial h for the difference quotient. Must be > 0; default = 1E-2. [in] max_steps The maximum number of Richardson extrapolation steps to be performed. Must be > 0; default = 10.
Definition at line 505 of file auto_derive.hpp.
References FEAT::Analytic::AutoDerive< Function_, DataType_ >::_init_grad_h, FEAT::Analytic::AutoDerive< Function_, DataType_ >::_max_grad_steps, and XASSERTM.
◆ config_hess_extrapol()
|
inline |
Configures the Hessian extrapolation scheme.
This function configures the Richardson extrapolation scheme for the evaluation of function hessians (second-order derivatives).
- Parameters
-
[in] initial_h The initial h for the difference quotient. Must be > 0; default = 1E-2. [in] max_steps The maximum number of Richardson extrapolation steps to be performed. Must be > 0; default = 10.
Definition at line 526 of file auto_derive.hpp.
References FEAT::Analytic::AutoDerive< Function_, DataType_ >::_init_hess_h, FEAT::Analytic::AutoDerive< Function_, DataType_ >::_max_hess_steps, and XASSERTM.
Member Data Documentation
◆ _init_grad_h
|
protected |
initial h for gradient extrapolation
Definition at line 465 of file auto_derive.hpp.
Referenced by FEAT::Analytic::AutoDerive< Function_, DataType_ >::config_grad_extrapol().
◆ _init_hess_h
|
protected |
initial h for hessian extrapolation
Definition at line 467 of file auto_derive.hpp.
Referenced by FEAT::Analytic::AutoDerive< Function_, DataType_ >::config_hess_extrapol().
◆ _max_grad_steps
|
protected |
maximum number of gradient extrapolation steps
Definition at line 461 of file auto_derive.hpp.
Referenced by FEAT::Analytic::AutoDerive< Function_, DataType_ >::config_grad_extrapol().
◆ _max_hess_steps
|
protected |
maximum number of hessian extrapolation steps
Definition at line 463 of file auto_derive.hpp.
Referenced by FEAT::Analytic::AutoDerive< Function_, DataType_ >::config_hess_extrapol().
◆ can_grad
|
staticconstexpr |
we provide function gradients
Definition at line 106 of file auto_derive.hpp.
◆ can_hess
|
staticconstexpr |
we provide function hessiants
Definition at line 108 of file auto_derive.hpp.
◆ can_value
|
staticconstexpr |
our base class provides function values
Definition at line 104 of file auto_derive.hpp.
◆ domain_dim
|
staticconstexpr |
the input function must support value computation
specify our domain dimension
Definition at line 98 of file auto_derive.hpp.
The documentation for this class was generated from the following file:
- kernel/analytic/auto_derive.hpp
