Flood::GradientDescent Class Reference
#include <GradientDescent.h>
Public Member Functions | |
| GradientDescent (ObjectiveFunctional *) | |
| GradientDescent (void) | |
| virtual | ~GradientDescent (void) |
| void | set_reserve_all_training_history (bool) |
| Vector< double > | calculate_training_direction (const Vector< double > &) |
| void | train (void) |
| void | resize_training_history (int) |
| std::string | get_training_history_XML (bool) |
| std::string | to_XML (bool) |
| void | load (const char *) |
Detailed Description
This concrete class represents the gradient descent training algorithm for an objective functional of a multilayer perceptron.Definition at line 29 of file GradientDescent.h.
Constructor & Destructor Documentation
| Flood::GradientDescent::GradientDescent | ( | ObjectiveFunctional * | new_objective_functional_pointer | ) | [explicit] |
General constructor. It creates a gradient descent training algorithm object associated to an objective functional object. It also initializes the class members to their default values:
Training operators:
- Training rate method = Brent;
Training parameters:
- First training rate: 1.0.
- Bracketing factor: 2.0.
- Tolerance in training rate: 1.0e-3.
Stopping criteria:
- Evaluation goal: -1.0e99.
- Gradient norm goal: 0.0.
- Maximum training time: 1.0e6.
- Maximum number of epochs: 100.
User stuff:
- Warning training rate: 1.0e6.
- Error training rate: 1.0e12.
- Display: true.
- Display period: 25.
- Parameters:
-
new_objective_functional_pointer Pointer to an objective functional object.
Definition at line 67 of file GradientDescent.cpp.
| Flood::GradientDescent::GradientDescent | ( | void | ) | [explicit] |
Default constructor. It creates a gradient descent training algorithm object not associated to any objective functional object. It also initializes the class members to their default values:
Training operators:
- Training rate method = Brent;
Training parameters:
- First training rate: 1.0.
- Bracketing factor: 2.0.
- Tolerance in training rate: 1.0e-3.
Stopping criteria:
- Evaluation goal: -1.0e99.
- Gradient norm goal: 0.0.
- Maximum training time: 1.0e6.
- Maximum number of epochs: 100.
User stuff:
- Warning training rate: 1.0e6.
- Error training rate: 1.0e12.
- Display: true.
- Display period: 25.
Definition at line 107 of file GradientDescent.cpp.
| Flood::GradientDescent::~GradientDescent | ( | void | ) | [virtual] |
Member Function Documentation
| std::string Flood::GradientDescent::get_training_history_XML | ( | bool | show_declaration | ) | [virtual] |
This method returns a string with the history of the training process in XML-type format. Only that training history variables which have been reserved are included here.
- Parameters:
-
show_declaration True if an XML-type declaration is to be included at the beginning.
Reimplemented from Flood::TrainingAlgorithm.
Definition at line 647 of file GradientDescent.cpp.
| void Flood::GradientDescent::load | ( | const char * | filename | ) | [virtual] |
This method loads a conjugate gradient object from an XML-type file. Please mind about the file format, wich is specified in the User's Guide.
- Parameters:
-
filename Name of conjugate gradient XML-type.
Reimplemented from Flood::TrainingAlgorithm.
Definition at line 730 of file GradientDescent.cpp.
| void Flood::GradientDescent::resize_training_history | ( | int | new_size | ) | [virtual] |
This method resizes the vectors or matrices containing training history information to a new size:
- Elapsed time history vector.
- Free parameters history matrix.
- Free parameters norm history vector.
- Evaluation history vector.
- Gradient history matrix.
- Gradient norm history vector.
- Training direction history matrix.
- Training direction norm history vector
- Training rate history vector.
- Parameters:
-
new_size Size of training history.
Reimplemented from Flood::TrainingAlgorithm.
Definition at line 1270 of file GradientDescent.cpp.
| void Flood::GradientDescent::set_reserve_all_training_history | ( | bool | new_reserve_all_training_history | ) | [virtual] |
This method makes the training history of all variables to reseved or not in memory:
- Parameters.
- Parameters norm.
- Evaluation.
- Gradient.
- Gradient norm.
- Validation error.
- Training direction.
- Training direction norm.
- Training rate.
- Parameters:
-
new_reserve_all_training_history True if the training history of all variables is to be reserved, false otherwise.
Reimplemented from Flood::TrainingAlgorithm.
Definition at line 139 of file GradientDescent.cpp.
| std::string Flood::GradientDescent::to_XML | ( | bool | show_declaration | ) | [virtual] |
This method prints to the screen the training parameters, the stopping criteria and other user stuff concerning the gradient descent object.
- Parameters:
-
show_declaration True if a XML-type declaration is wanted, false otherwise.
Reimplemented from Flood::TrainingAlgorithm.
Definition at line 514 of file GradientDescent.cpp.
| void Flood::GradientDescent::train | ( | void | ) | [virtual] |
This method trains a multilayer perceptron with an associated objective functional according to the gradient descent algorithm. Training occurs according to the training operators, training parameters and stopping criteria.
Implements Flood::TrainingAlgorithm.
Definition at line 180 of file GradientDescent.cpp.
The documentation for this class was generated from the following files:
