STIR  6.2.0
Public Types | Public Member Functions | Protected Member Functions | List of all members
stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT > Class Template Reference

Test class for GeneralisedObjectiveFunction and GeneralisedPrior. More...

#include "stir/recon_buildblock/test/ObjectiveFunctionTests.h"

Inheritance diagram for stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >:
Inheritance graph
[legend]

Public Types

typedef ObjectiveFunctionT objective_function_type
 
typedef TargetT target_type
 

Public Member Functions

virtual Succeeded test_gradient (const std::string &test_name, ObjectiveFunctionT &objective_function, TargetT &target, const float eps, const bool full_gradient=true)
 Test the gradient of the objective function by comparing to the numerical gradient via perturbation. More...
 
virtual Succeeded test_Hessian (const std::string &test_name, ObjectiveFunctionT &objective_function, const TargetT &target, const float eps)
 Test the accumulate_Hessian_times_input of the objective function by comparing to the numerical result via perturbation. More...
 
virtual Succeeded test_Hessian_concavity (const std::string &test_name, ObjectiveFunctionT &objective_function, const TargetT &target, const float mult_factor=1.F)
 Test the Hessian of the objective function by testing the (mult_factor * x^T Hx > 0) condition. More...
 
- Public Member Functions inherited from stir::RunTests
 RunTests (const double tolerance=1E-4)
 Default constructor.
 
virtual ~RunTests ()
 Destructor, outputs a diagnostic message.
 
virtual void run_tests ()=0
 Function (to be overloaded) which does the actual tests. More...
 
bool is_everything_ok () const
 Returns if all checks were fine upto now.
 
int main_return_value () const
 Handy return value for a main() function. More...
 
void set_tolerance (const double tolerance)
 Set value used in floating point comparisons (see check_* functions)
 
double get_tolerance () const
 Get value used in floating point comparisons (see check_* functions)
 
bool check (const bool, const std::string &str="")
 Tests if true, str can be used to tell what you are testing. More...
 
template<class T1 , class T2 >
bool check_if_less (T1 a, T2 b, const std::string &str="")
 check if a<b
 
bool check_if_equal (const std::string &a, const std::string &b, const std::string &str="")
 
bool check_if_equal (const double a, const double b, const std::string &str="")
 
bool check_if_equal (const short a, const short b, const std::string &str="")
 
bool check_if_equal (const unsigned short a, const unsigned short b, const std::string &str="")
 
bool check_if_equal (const int a, const int b, const std::string &str="")
 
bool check_if_equal (const unsigned int a, const unsigned int b, const std::string &str="")
 
bool check_if_equal (const long a, const long b, const std::string &str="")
 
bool check_if_equal (const unsigned long a, const unsigned long b, const std::string &str="")
 
bool check_if_equal (const Bin &a, const Bin &b, const std::string &str="")
 
template<class T >
bool check_if_equal (const DetectionPosition< T > &a, const DetectionPosition< T > &b, const std::string &str="")
 
template<class T >
bool check_if_equal (const std::complex< T > a, const std::complex< T > b, const std::string &str="")
 check equality by calling check_if_equal on real and imaginary parts
 
template<class T >
bool check_if_equal (const VectorWithOffset< T > &t1, const VectorWithOffset< T > &t2, const std::string &str="")
 check equality by comparing ranges and calling check_if_equal on all elements
 
template<class T >
bool check_if_equal (const std::vector< T > &t1, const std::vector< T > &t2, const std::string &str="")
 check equality by comparing size and calling check_if_equal on all elements
 
bool check_if_equal (const ProjDataInMemory &t1, const ProjDataInMemory &t2, const std::string &str="")
 
template<int n>
bool check_if_equal (const IndexRange< n > &t1, const IndexRange< n > &t2, const std::string &str="")
 
template<int num_dimensions, class coordT >
bool check_if_equal (const BasicCoordinate< num_dimensions, coordT > &a, const BasicCoordinate< num_dimensions, coordT > &b, const std::string &str="")
 check equality by comparing norm(a-b) with tolerance
 
bool check_if_zero (const double a, const std::string &str="")
 
bool check_if_zero (const short a, const std::string &str="")
 
bool check_if_zero (const unsigned short a, const std::string &str="")
 
bool check_if_zero (const int a, const std::string &str="")
 
bool check_if_zero (const unsigned int a, const std::string &str="")
 
bool check_if_zero (const long a, const std::string &str="")
 
bool check_if_zero (const unsigned long a, const std::string &str="")
 
template<class T >
bool check_if_zero (const VectorWithOffset< T > &t, const std::string &str="")
 use check_if_zero on all elements
 
template<int num_dimensions, class coordT >
bool check_if_zero (const BasicCoordinate< num_dimensions, coordT > &a, const std::string &str="")
 compare norm with tolerance
 

Protected Member Functions

virtual shared_ptr< const TargetT > construct_increment (const TargetT &target, const float eps) const
 Construct small increment for target. More...
 
- Protected Member Functions inherited from stir::RunTests
template<class T >
bool check_if_equal_generic (const T &a, const T &b, const std::string &str)
 function that is called by some check_if_equal implementations. It just uses operator!=
 
template<class T >
bool check_if_zero_generic (T a, const std::string &str)
 function that is called by some check_if_zero implementations. It just uses operator!=
 

Additional Inherited Members

- Protected Attributes inherited from stir::RunTests
double tolerance
 tolerance for comparisons with real values
 
bool everything_ok
 variable storing current status
 

Detailed Description

template<class ObjectiveFunctionT, class TargetT>
class stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >

Test class for GeneralisedObjectiveFunction and GeneralisedPrior.

This contains some numerical tests to check gradient and Hessian calculations.

Note that the test only works if the objective function is well-defined. For example, if certain projections are non-zero, while the model estimates them to be zero, the Poisson objective function is in theory infinite. ObjectiveFunction uses some thresholds to try to avoid overflow, but if there are too many of these bins, the total objective function will become infinite. The numerical gradient then becomes ill-defined (even in voxels that do not contribute to these bins).

Member Function Documentation

◆ test_gradient()

template<class ObjectiveFunctionT, class TargetT>
Succeeded stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >::test_gradient ( const std::string &  test_name,
ObjectiveFunctionT &  objective_function,
TargetT &  target,
const float  eps,
const bool  full_gradient = true 
)
virtual

Test the gradient of the objective function by comparing to the numerical gradient via perturbation.

If full_gradient=true, all elements in the gradient are tested (using single-element increments). This is slow. Otherwise, the test checks that $ G dx \approx F(x+dx) - F(x) $. dx is computed via construct_increment().

Note: target is non-const, as the code will add/subtract eps, but the actual values are not modified after the test exits.

◆ test_Hessian()

template<class ObjectiveFunctionT, class TargetT>
Succeeded stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >::test_Hessian ( const std::string &  test_name,
ObjectiveFunctionT &  objective_function,
const TargetT &  target,
const float  eps 
)
virtual

Test the accumulate_Hessian_times_input of the objective function by comparing to the numerical result via perturbation.

This test checks that $ H dx \approx G(x+dx) - G(x) $. dx is computed via construct_increment().

◆ test_Hessian_concavity()

template<class ObjectiveFunctionT, class TargetT>
Succeeded stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >::test_Hessian_concavity ( const std::string &  test_name,
ObjectiveFunctionT &  objective_function,
const TargetT &  target,
const float  mult_factor = 1.F 
)
virtual

Test the Hessian of the objective function by testing the (mult_factor * x^T Hx > 0) condition.

setup images

Compute H x

Compute dot(x,(H x))

◆ construct_increment()

template<class ObjectiveFunctionT , class TargetT>
shared_ptr< const TargetT > stir::ObjectiveFunctionTests< ObjectiveFunctionT, TargetT >::construct_increment ( const TargetT &  target,
const float  eps 
) const
protectedvirtual

Construct small increment for target.

Result is eps*(target / target.find_max() + 0.5), i.e. it is always positive (if target is non-negative).


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