keel.Algorithms.Shared.ClassicalOptim

Class ConjGradNN

java.lang.Object

keel.Algorithms.Shared.ClassicalOptim.ConjGradNN

public class ConjGradNN
extends java.lang.Object

Optimized Classificator/Model by Conjugated Gradient. Also this class is a container for a perceptron neural network and implements the training methods: * Conjugated Gradient: conjugatedGradient. * Descendent Gradient: descentGradient. Input-Layer Hidden Layer-i x nLayers Output-Layer - | I H | I H - | I H O | nInputs | I H O | nOutputs | I H O | | I H O | | I H - | I H -

Since:

JDK1.4

Version:

1.0

Author:

Written by Luciano Sánchez (University of Oviedo) 27/02/2004, Modified by Enrique A. de la Cal (University of Oviedo) 13/12/2008

Constructor Summary

Constructors

Constructor and Description
ConjGradNN(int[] vNelement, double[][] vInput, double[][] vOutput, Randomize pr) Constructor for a perceptron neural network from its basic elements.

Method Summary

Modifier and Type	Method and Description
void	changeWeights(double x) Updates the matrix of weights with the addition of random valued in the range [-x,x].
double	conjugatedGradient(FUN f, double TOL_ERR, double MIN_DELTAGC, int MAX_ITER) Returns the mean square error of the output perceptron calculated with Conjugated Gradient training algorithm.
void	debugOutput() Prints to standard output the main information about the training algorithm run: -the matrix of weights -the original input examples (not scaled) -the original obtained output (not scaled) -the original expected output (not scaled)
double	descentGradient(FUN f, double TOL_ERR, double MIN_DELTAGC, int MAX_ITER) Returns the mean square error of the output perceptron calculated with Descendent Gradient training algorithm.
double	f_denormalized(double[][][] x, double[] FACTOR) Returns the denormalized mean square error of a perceptron with weights x for all the examples Input
double	f(double[][][] x) Returns the mean square error of a perceptron with weights x for all the examples Input
void	getWeights(double[] p) Returns a copy of weights in vector p.
java.lang.String	getWeightsAsString() Return the weights as a String to print them
double[]	nn(double[] x, double[][][] W) Calculates the output of a perceptron with weights W for input x
void	randomWeights(double x) Initializes the matrix of weights with random valued in the range [-x,x].
void	sample(double[][][] x) Prints to standard output N-tier Neural Network x
void	scale() Scales the input examples values and expected output valued
void	setWeights(double[] p) Copy the weights contained in p to the weights matrix.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

ConjGradNN

public ConjGradNN(int[] vNelement,
                  double[][] vInput,
                  double[][] vOutput,
                  Randomize pr)

Constructor for a perceptron neural network from its basic elements.

Parameters:

vNelement - topology. Number of neurons by hidden layer.

vInput - input examples.

vOutput - expected outputs.

pr - Random generator.

Method Detail

sample

public void sample(double[][][] x)

Prints to standard output N-tier Neural Network x

Parameters:

x - the example to print

nn

public double[] nn(double[] x,
                   double[][][] W)

Calculates the output of a perceptron with weights W for input x

Parameters:

x - the example to give the perceptron

W - the weights of the perceptron

Returns:

the output of perceptron with weights W for input x

f

public double f(double[][][] x)

Returns the mean square error of a perceptron with weights x for all the examples Input

Parameters:

x - the example to give the perceptron

Returns:

the mean square error of the perceptron x output respect the expected output Output

f_denormalized

public double f_denormalized(double[][][] x,
                             double[] FACTOR)

Returns the denormalized mean square error of a perceptron with weights x for all the examples Input

Parameters:

x - the example to give the perceptron.

FACTOR - a vector with difference between the max and min value for each output

Returns:

the denormalized mean square error of the perceptron x output respect the expected output Output

scale

public void scale()

Scales the input examples values and expected output valued

conjugatedGradient

public double conjugatedGradient(FUN f,
                                 double TOL_ERR,
                                 double MIN_DELTAGC,
                                 int MAX_ITER)

Returns the mean square error of the output perceptron calculated with Conjugated Gradient training algorithm.

Parameters:

f - the error function.

TOL_ERR - the stop error.

MIN_DELTAGC - is not used.

MAX_ITER - number of maximum iteratations.

Returns:

the mean square error of the output perceptron calculated with Conjugated Gradient training algorithm.

descentGradient

public double descentGradient(FUN f,
                              double TOL_ERR,
                              double MIN_DELTAGC,
                              int MAX_ITER)

Returns the mean square error of the output perceptron calculated with Descendent Gradient training algorithm.

Parameters:

f - the error function.

TOL_ERR - the stop error.

MIN_DELTAGC - is not used.

MAX_ITER - number of maximum iteratations.

Returns:

the mean square error of the output perceptron calculated with Descendent Gradient training algorithm.

randomWeights

public void randomWeights(double x)

Initializes the matrix of weights with random valued in the range [-x,x].

Parameters:

x - the lower/upper limit for random values.

changeWeights

public void changeWeights(double x)

Updates the matrix of weights with the addition of random valued in the range [-x,x].

Parameters:

x - the lower/upper limit for random values.

debugOutput

public void debugOutput()

Prints to standard output the main information about the training algorithm run: -the matrix of weights -the original input examples (not scaled) -the original obtained output (not scaled) -the original expected output (not scaled)

getWeights

public void getWeights(double[] p)

Returns a copy of weights in vector p.

Parameters:

p - an output parameter to obtaing a copy of weights.

setWeights

public void setWeights(double[] p)

Copy the weights contained in p to the weights matrix.

Parameters:

p - an output parameter to obtaing a copy of weights.

getWeightsAsString

public java.lang.String getWeightsAsString()

Return the weights as a String to print them

Returns:

the weights in String format