Class NNeco

java.lang.Object
  |
  +--NNeco

public class NNeco

extends java.lang.Object

NNeco.java
Title : Learning Industry with ANN based GenFirms
Description : A simple back propagation feed forward neural network adapted for being driven by another user class. It has only one hidden layer.
Adapted with corrections from the Neural class by Mark Watson, 1997, Intelligent Java Applications, Morgan Kaufmann, SF:CA. My modifications are based on T. Masters, 1993, Practical Neural Recipes in C++, Academic Press, San Diego: CA. The later is a very nice reference on ANNs.
Copyright : Copyright (c) Murat Yildizoglu August 2001
Permission to use, copy, modify, and distribute this
software and its documentation for NON-COMMERCIAL purposes
and without fee is hereby granted provided that this
copyright notice appears in all copies.
*************************************************************

Version:

1.0 August 2001

Author:

Murat Yildizoglu

Field Summary

protected double[]	Hidden
protected double[]	hidden_errors
double[]	Inputs
protected int	NumHidden
protected int	NumInputs
protected int	NumOutputs
protected int	NumTraining
protected double[]	output_errors
double[]	Outputs
protected double[][]	W1
protected double[][]	W2

Constructor Summary

NNeco()
The default constructor.

NNeco(int i, int h, int o)
The constructor that should be used for creating a new network

Method Summary

void	forwardPass() Calculates outputs with actual weights and input vector.
double[]	getOutputs() This method returns the actual values of the outputs resulting from the last forward pass.
double	getTotalOutputError() Returns the total error in the last training sesssion
double[]	infer(double[] inputs) This is the method for obtaining predictions from the network.
void	randomizeWeights() Puts randomly chosen values to all weights
void	setInputs(double[] inputs) This method uses the provided one-dimensional array for setting the values of the inputs
protected double	Sigmoid(double x) The sigmoid function used for activation
protected double	SigmoidP(double x) The derivative of the sigmoid function.
void	Train(double[][] ins, double[][] outs, int numEpoch, int numCases, int winSize) Takes the data in a more standard two dimensional table and adapts it for using in the real train method.
double	Train(double[] ins, double[] outs, int num_cases) This method is called by the preceding Train method, it should not be directly called.

Methods inherited from class java.lang.Object

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

Field Detail

NumInputs

protected int NumInputs

NumHidden

protected int NumHidden

NumOutputs

protected int NumOutputs

NumTraining

protected int NumTraining

Inputs

public double[] Inputs

Hidden

protected double[] Hidden

Outputs

public double[] Outputs

W1

protected double[][] W1

W2

protected double[][] W2

output_errors

protected double[] output_errors

hidden_errors

protected double[] hidden_errors

Constructor Detail

NNeco

public NNeco()

The default constructor. Not very useful.

NNeco

public NNeco(int i,
             int h,
             int o)

The constructor that should be used for creating a new network

Parameters:

i - number of inputs

h - number of hidden nodes on the hidden layer

o - number of outputs

Method Detail

randomizeWeights

public void randomizeWeights()

Puts randomly chosen values to all weights

forwardPass

public void forwardPass()

Calculates outputs with actual weights and input vector. The inputs must be set with the Train or setInputs methods.

Train

public void Train(double[][] ins,
                  double[][] outs,
                  int numEpoch,
                  int numCases,
                  int winSize)

Takes the data in a more standard two dimensional table and adapts it for using in the real train method.

Parameters:

ins - two-dimensional table where each case is a line of input values

outs - two-dimensional table where each case is a line of output values

numEpoch - the number of runs in each epoch

numCases - the size of the epoch (number of training cases)

winSize - the number of observations to use from the end of the tables. 0 means all observations.

Train

public double Train(double[] ins,
                    double[] outs,
                    int num_cases)

This method is called by the preceding Train method, it should not be directly called.

infer

public double[] infer(double[] inputs)

This is the method for obtaining predictions from the network. It gives the provided values to the inputs, runs through the network (forward pass) and returns an array containing the values of all outputs.

Parameters:

inputs - a one-dimensional array that contains the vector of input values

getOutputs

public double[] getOutputs()

This method returns the actual values of the outputs resulting from the last forward pass.

getTotalOutputError

public double getTotalOutputError()

Returns the total error in the last training sesssion

setInputs

public void setInputs(double[] inputs)

This method uses the provided one-dimensional array for setting the values of the inputs

Parameters:

inputs - a one-dimensional array of iputs values.

Sigmoid

protected double Sigmoid(double x)

The sigmoid function used for activation

Parameters:

x -

SigmoidP

protected double SigmoidP(double x)

The derivative of the sigmoid function. It is used during the bacward propagation of errors

Parameters:

x -