Main Page   Compound List   File List   Compound Members  

PerceptronNeuron Class Reference

#include <PerceptronNeuron.h>

List of all members.

Public Member Functions

 PerceptronNeuron (unsigned int in_layer_num)
double getDelta (void)
void setDelta (double newval)
double getTheta (void)
void setTheta (double newval)
double getThetaDiff (void)
void initializeWeightings (unsigned int succ_count)
void resetWeights (void)
void resetWeightDiffs (void)
void resetDiffs (void)
void postprocessWeight (PerceptronNeuron *succ, double epsilon, double weight_decay, double momterm)
void postprocessTheta (double epsilon, double weight_decay, double momterm)
void update (void)

Public Attributes

unsigned int num
double input
double output
vector< double > weight
vector< double > weight_diff

Protected Attributes

double delta
double theta
double theta_diff
double theta_diff_last
vector< double > weight_diff_last

Detailed Description

Represent one neuron within a layer of the neural network.

Constructor & Destructor Documentation

PerceptronNeuron::PerceptronNeuron unsigned int    in_layer_num

Main constructor.

in_layer_num number of neuron within its layer.

Member Function Documentation

double PerceptronNeuron::getDelta void    [inline]

Getter function for the delta variable.

double PerceptronNeuron::getTheta void    [inline]

Getter function for theta parameter.

double PerceptronNeuron::getThetaDiff void    [inline]

Getter for the difference calculated for the theta parameter.

void PerceptronNeuron::initializeWeightings unsigned int    succ_count

Initialize the weightings vectors used for each neuron to zero.

succ_count number of successor neurons to this neuron.

void PerceptronNeuron::postprocessTheta double    epsilon,
double    weight_decay,
double    momterm

Theta postprocess algorithm. Assign the theta differences to this neuron.

epsilon Learning parameter.
weight_decay Weight decay factor.
momterm Momentum term factor.

void PerceptronNeuron::postprocessWeight PerceptronNeuron *    succ,
double    epsilon,
double    weight_decay,
double    momterm

Weighting postprocess algorithm. Assign the weight differences to this neuron. Note that the assignment is done by adding the new delta value to the current one. Hence, both batch- and online learning can use this function. For online-learning the values must be resetted after each update using the resetWeightDiffs method. Also, only one weighting difference is calculated at a time. The computation of all differences is scheduled by PerceptronLayer::postprocess.

succ Successor neuron to the current neuron.
epsilon Learning parameter.
weight_decay Weight decay factor.
momterm Momentum term factor.
See also:
resetWeightDiffs ()

void PerceptronNeuron::resetDiffs void   

Reset all learned differences.

void PerceptronNeuron::resetWeightDiffs void   

Reset all weight deltas to zero.

void PerceptronNeuron::resetWeights void   

Reset all weightings to zero.

void PerceptronNeuron::setDelta double    newval [inline]

Setter function for the delta variable.

newval New value to be assigned to delta.

void PerceptronNeuron::setTheta double    newval [inline]

Setter function for theta parameter.

void PerceptronNeuron::update void   

Update algorithm. Update weightings and theta parameter by the values calculated in the postprocess step.

Member Data Documentation

double PerceptronNeuron::delta [protected]

Errorsignal, computed by the backpropagation algorithm. Used to compute the deltas for both the weightings and the sensitivity parameter

double PerceptronNeuron::input

Input signal level for this node, .

unsigned int PerceptronNeuron::num

Counter within the current layer (top = 0)

double PerceptronNeuron::output

Output signal level for this node, .

double PerceptronNeuron::theta [protected]

Sensitivity parameter to the activation function

double PerceptronNeuron::theta_diff [protected]

Delta-value for the sensitivity, calculated by the Processing algorithm, used to update theta within the Update algorithm

double PerceptronNeuron::theta_diff_last [protected]

The previous theta difference, used for momentum term calculation.

vector<double> PerceptronNeuron::weight

Weightings to neurons in the next layer. Hence, the size of the vector must equal the number of neurons in the next layer

vector<double> PerceptronNeuron::weight_diff

Delta-value for the individual weightings the node has to its successor nodes. Computed by the Processing algorithm

Note this is within the public space due to the from-file constructor of PerceptronNetwork, which push_back's zeroes here.

vector<double> PerceptronNeuron::weight_diff_last [protected]

The previous weight differences, used for momentum term.

The documentation for this class was generated from the following files:
Generated on Sun Mar 2 21:35:50 2003 for libperceptronnetwork by doxygen1.3-rc3