DefectTrainer Class Reference

Class dedicated to training mode. More...

Inheritance diagram for DefectTrainer:

List of all members.

Public Member Functions
	DefectTrainer (const DefectixOptions::Options &options)
	Standard constructor.
const errorCodes	exec ()
	Main function.
const errorCodes	createNetworks ()
	create networks recursively
Private Member Functions
errorCodes	createVectors (gsl_matrix *input, gsl_matrix *output, std::map< int, gsl_matrix * > &inputSrc, std::map< int, gsl_matrix * > &outputSrc, FitsManager &mgr)
	create vectors needed for training
void	createKLBase (gsl_matrix *inputV)
	compute or load a KL-Base and compute PCA
const errorCodes	createNetwork (gsl_matrix *inputV, gsl_matrix *outputV)
	create a network and train it
Private Attributes
t_network *	_l0
	Neural network.
int	_nbPC
	Number of principal components.
int	_nbVectors
	Number of vectors of the training base.
unsigned	_maxIter
	Maximum number of iterations.
double	_energyTreshold
	Network energy treshold.
std::string	_evecFile
	Name of the file containing the KL-Base.

---

Detailed Description

Class dedicated to training mode.

The class is the training part of the project.

---

Constructor & Destructor Documentation

DefectTrainer::DefectTrainer (  const DefectixOptions::Options &  options  )

Standard constructor. The constructor calls Defectix constructor and then initialises private members with 'options' values. Parameters: options  Class containing every parameters set by the user (or default)

---

Member Function Documentation

void DefectTrainer::createKLBase (  gsl_matrix *  inputV  )  [private]

compute or load a KL-Base and compute PCA The functions computes PCA. If a file DefectTrainer::_evecFile exists, it is loaded as the KL-Base. If no file exists, the base is computed first. Parameters: inputV  input vector to transform on the KL-Base

const errorCodes DefectTrainer::createNetwork (  gsl_matrix *  inputV, gsl_matrix *  outputV )  [private]

create a network and train it The function creates a network accordingly to input and output vectors. The network is then trained using rprop algorithm. Parameters: inputV  input vector (obtained with createVectors()) outputV  output vector (obtained with createVectors())

const errorCodes DefectTrainer::createNetworks (   )

create networks recursively The functions creates networks and trains them recursively

errorCodes DefectTrainer::createVectors (  gsl_matrix *  input, gsl_matrix *  output, std::map< int, gsl_matrix * > &  inputSrc, std::map< int, gsl_matrix * > &  outputSrc, FitsManager &  mgr )  [private]

create vectors needed for training The functions reads images and masks and create input and output vectors for training. Parameters: input  neural network input vector to fill output  neural network output vector to fill inputSrc  image source list outputSrc  mask source list mgr  image manager to handle lists

const errorCodes DefectTrainer::exec (   )  [virtual]

Main function. The function starts the training Implements Defectix.

---

Member Data Documentation

double DefectTrainer::_energyTreshold [private]

Network energy treshold. _energyTreshold is the energy limit the network must reach to stop converging.

std::string DefectTrainer::_evecFile [private]

Name of the file containing the KL-Base. _evecFile is the name of the KL-Base stored as a gsl_matrix.

t_network * DefectTrainer::_l0 [private]

Neural network. _l0 is the neural network used in DefectTrainer::createNetwork()

unsigned DefectTrainer::_maxIter [private]

Maximum number of iterations. _maxIter is the maximum number of iterations the network is allowed to do without any energy decrease. This parameter avoid an infinite loop if _energyTreshold cannot be reached.

int DefectTrainer::_nbPC [private]

Number of principal components. _nbPC is the number of principal components to use once the KL-Base is computed.

int DefectTrainer::_nbVectors [private]

Number of vectors of the training base. _nbVectors is the number of vectors create in the input and output sets to train the network

---

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

• trainer.dox

---