FeedForwardNetwork Class Reference

A Feed Forward Network. More...

#include <Gist/FFN.H>

Collaboration diagram for FeedForwardNetwork:

Constructors, assignment, and Destructors

	FeedForwardNetwork ()
	Constructor.
	~FeedForwardNetwork ()
	Destructor.
void	init (std::string wh_file, std::string wo_file, int inunits, int hidunits, int outunits, double lrate, double mrate)
void	init (int inunits, int hidunits, int outunits, double lrate, double mrate)
void	init (Image< double > wh, Image< double > wo, double lrate, double mrate)
	Initialize a two-weight-layer network with Image<double> weights.
void	init3L (std::string wh1_file, std::string wh2_file, std::string wo_file, int inunits, int hid1units, int hid2units, int outunits, double lrate, double mrate)
void	init3L (int inunits, int hid1units, int hid2units, int outunits, double lrate, double mrate)
void	init3L (Image< double > wh, Image< double > wh2, Image< double > wo, double lrate, double mrate)
	Initialize a three-weight-layer network with Image<double> weights.
Image< double >	run (Image< double > input)
	Run a two-weight-layer network with passed in input.
Image< double >	run3L (Image< double > input)
	Run a three-weight-layer network with passed in input.
void	backprop (Image< double > target)
void	backprop3L (Image< double > target)
void	write (std::string wh_file, std::string wo_file)
	Write the two-weight-layer network weights to a specified file.
void	write3L (std::string wh1_file, std::string wh2_file, std::string wo_file)
	Write the three-weight-layer network weights to a specified file.
void	setLearningRate (double newLR)
	Set Learning rate - for simulated annealing.
Image< double >	getOutput ()
	get the most recent output

Detailed Description

A Feed Forward Network.

Definition at line 54 of file FFN.H.

Constructor & Destructor Documentation

FeedForwardNetwork::FeedForwardNetwork ( )

Constructor.

This implements a simple feedforward network with backpropagation. Supports 2 or 3-weight-layer networks through separate function calls NOTE: please allow the class to create the initial weight files (use the init function that does not ask for weight files). Also user should add an extra input for bias (set it to 1.0)

Definition at line 57 of file FFN.C.

FeedForwardNetwork::~FeedForwardNetwork ( )

Destructor.

Definition at line 61 of file FFN.C.

Member Function Documentation

void FeedForwardNetwork::backprop ( Image< double > target )

backpropagate the error in the two-weight-layer network with passed in target output (note: call run first)

Definition at line 468 of file FFN.C.

References Image< T >::getDims(), Image< T >::getSize(), matrixMult(), transpose(), and ZEROS.

void FeedForwardNetwork::backprop3L ( Image< double > target )

backpropagate the error in the three-weight-layer network with passed in target output (note: call run3L first)

Definition at line 498 of file FFN.C.

References Image< T >::getDims(), Image< T >::getSize(), matrixMult(), transpose(), and ZEROS.

Image< double > FeedForwardNetwork::getOutput ( ) [inline]

get the most recent output

Definition at line 167 of file FFN.H.

void FeedForwardNetwork::init	(	Image< double >	wh,
		Image< double >	wo,
		double	lrate,
		double	mrate
	)

Initialize a two-weight-layer network with Image<double> weights.

Definition at line 178 of file FFN.C.

References ASSERT, Image< T >::getHeight(), Image< T >::getWidth(), Image< T >::resize(), and Image< T >::setVal().

void FeedForwardNetwork::init	(	int	inunits,
		int	hidunits,
		int	outunits,
		double	lrate,
		double	mrate
	)

Initialize a two-weight-layer network with small random weights momentum is unimplemented

Definition at line 82 of file FFN.C.

void FeedForwardNetwork::init	(	std::string	wh_file,
		std::string	wo_file,
		int	inunits,
		int	hidunits,
		int	outunits,
		double	lrate,
		double	mrate
	)

Initialize a two-weight-layer network with saved weights momentum is unimplemented

Definition at line 90 of file FFN.C.

References Image< T >::beginw(), NO_INIT, Image< T >::resize(), Image< T >::setVal(), and transpose().

void FeedForwardNetwork::init3L	(	Image< double >	wh,
		Image< double >	wh2,
		Image< double >	wo,
		double	lrate,
		double	mrate
	)

Initialize a three-weight-layer network with Image<double> weights.

Definition at line 354 of file FFN.C.

References ASSERT, Image< T >::getHeight(), and Image< T >::getWidth().

void FeedForwardNetwork::init3L	(	int	inunits,
		int	hid1units,
		int	hid2units,
		int	outunits,
		double	lrate,
		double	mrate
	)

Initialize a two-weight-layer network with small random weights momentum is unimplemented

Definition at line 230 of file FFN.C.

void FeedForwardNetwork::init3L	(	std::string	wh1_file,
		std::string	wh2_file,
		std::string	wo_file,
		int	inunits,
		int	hid1units,
		int	hid2units,
		int	outunits,
		double	lrate,
		double	mrate
	)