mimas::nn_mlp_tool Class Reference

Multi-layered perceptron (MLP) tool. More...

#include <nn_mlp_tool.h>

Inheritance diagram for mimas::nn_mlp_tool:

[legend]Collaboration diagram for mimas::nn_mlp_tool:


Public Types
typedef boost::numeric::ublas::matrix< double >	Matrix
typedef boost::numeric::ublas::vector< double >	Vector
Public Member Functions
	nn_mlp_tool (void)
	constructor
	~nn_mlp_tool (void)
	destructor
void	init (int nodes,...) throw (mimasexception)
	initialise network with number of nodes in each layer. List must be NULL terminated!
void	clear (void)
	delete the whole network
void	showConfig (void)
	show network configuration
Matrix	feedForward (Matrix &pat) throw (mimasexception)
	feedforward patterns and return a matrix containing the results.
void	save (const char *fn) throw (mimasexception)
	save neural network weights to a file. Note that transfer function information are not saved!
void	load (const char *fn) throw (mimasexception)
	load neural network from a file. Note that transfer function information are not loaded!
void	setIterations (int i) throw (mimasexception)
	set number of iterations to train for
int	getIterations (void)
	get number of iterations to train for
Matrix	trainOnline (Matrix &pat, double learningRate=0.3, double momentum=0.05, double weightDecay=0.0) throw (mimasexception)
	online training of network on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.
Matrix	trainBatch (Matrix &pat, double learningRate=0.3, double momentum=0.05, double weightDecay=0.0) throw (mimasexception)
	batch training of network on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.
Matrix	trainQprop (Matrix &pat, double learningRate=0.0005, double momentum=0.05, double maxFactor=2.0) throw (mimasexception)
	batch training using Quickprop on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.
Matrix	trainRprop (Matrix &pat, double etapos=1.2, double etaneg=0.5) throw (mimasexception)
	batch training using RProp on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.
void	setTransferFuncLayer (ptrToFuncType newfunc, int layer) throw (mimasexception)
	set transfer function for layer. Note that first hidden layer is layer 2. Must be set together with setTransferFuncDerivLayer()
void	setTransferFuncDerivLayer (ptrToFuncType newfunc, int layer) throw (mimasexception)
	set derivative of transfer function for layer. Note that first hidden layer is layer 2. Must be set together with setTransferFuncLayer()
virtual double	transferFunc (double val)
	a transfer function that is overwritable by the user
virtual double	transferFuncDeriv (double val)
	the derivative of a transfer function that is overwritable by the user
double	transferFuncTanh (double val)
	the hyperbolic tangent transfer function
double	transferFuncDerivTanh (double val)
	the derivative of hyperbolic tangent the transfer function
double	transferFuncLinear (double val)
	the linear transfer function
double	transferFuncDerivLinear (double val)
	the derivative of linear the transfer function
void	setDebug (bool val)
	set/unset debugging flag. When set, training informaiton will be displayed

Detailed Description

Multi-layered perceptron (MLP) tool.

Author:: Bala Amavasai (bala@amavasai.org)

Date:

Tue Sep 24 13:19:08 BST 2002

Header: /cvs/mimas2/include/nn_mlp_tool.h,v 1.1.1.1 2005/08/09 15:37:45 engmb Exp

Definition at line 28 of file nn_mlp_tool.h.

Member Typedef Documentation

typedef boost::numeric::ublas::matrix< double > mimas::nn_mlp_tool::Matrix

Definition at line 31 of file nn_mlp_tool.h.

typedef boost::numeric::ublas::vector< double > mimas::nn_mlp_tool::Vector

Definition at line 32 of file nn_mlp_tool.h.

Constructor & Destructor Documentation

mimas::nn_mlp_tool::nn_mlp_tool ( void )

constructor

mimas::nn_mlp_tool::~nn_mlp_tool ( void )

destructor

Member Function Documentation

void mimas::nn_mlp_tool::init	(	int	nodes,
			...
	)			throw (mimasexception)

initialise network with number of nodes in each layer. List must be NULL terminated!

void mimas::nn_mlp_tool::clear ( void )

delete the whole network

void mimas::nn_mlp_tool::showConfig ( void )

show network configuration

Matrix mimas::nn_mlp_tool::feedForward ( Matrix & pat ) throw (mimasexception)

feedforward patterns and return a matrix containing the results.

void mimas::nn_mlp_tool::save ( const char * fn ) throw (mimasexception)

save neural network weights to a file. Note that transfer function information are not saved!

void mimas::nn_mlp_tool::load ( const char * fn ) throw (mimasexception)

load neural network from a file. Note that transfer function information are not loaded!

void mimas::nn_mlp_tool::setIterations ( int i ) throw (mimasexception)

set number of iterations to train for

int mimas::nn_mlp_tool::getIterations ( void )

get number of iterations to train for

Matrix mimas::nn_mlp_tool::trainOnline	(	Matrix &	pat,
		double	learningRate = `0.3`,
		double	momentum = `0.05`,
		double	weightDecay = `0.0`
	)			throw (mimasexception)

online training of network on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.

Matrix mimas::nn_mlp_tool::trainBatch	(	Matrix &	pat,
		double	learningRate = `0.3`,
		double	momentum = `0.05`,
		double	weightDecay = `0.0`
	)			throw (mimasexception)

batch training of network on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.

Matrix mimas::nn_mlp_tool::trainQprop	(	Matrix &	pat,
		double	learningRate = `0.0005`,
		double	momentum = `0.05`,
		double	maxFactor = `2.0`
	)			throw (mimasexception)

batch training using Quickprop on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.

Matrix mimas::nn_mlp_tool::trainRprop	(	Matrix &	pat,
		double	etapos = `1.2`,
		double	etaneg = `0.5`
	)			throw (mimasexception)

batch training using RProp on input/output dataset pat. Note that the number of columns that are the input patters and target patters are determined by the number of inputs and outputs of the network during the init() call. Returns 2-column matrix of sum-squared-error (SSE) values.