BPnnet Class Reference

describes structure of a 3 layer back prop neural net More...

#include <BPnnet/BPnnet.H>

Collaboration diagram for BPnnet:

[legend]

List of all members.

Public Member Functions
	BPnnet (const int numInput, const int numHidden, const KnowledgeBase *kb)
	Constructor.
	~BPnnet ()
	Destructor.
void	randomizeWeights ()
	Initialize all weights to (small) random values.
void	normalizeWeights ()
	Normalize weights.
double	train (const Image< float > &in, const SimpleVisualObject &target, const double learnRate)
	Do one training iteration.
bool	recognize (const Image< float > &in, SimpleVisualObject &vo)
	Attempt to recognize given jet as a certain visual object.
bool	save (const char *filename) const
	Store 2 matrices of weights to the file "filename".
bool	load (const char *filename)
	Assign all weights based on data stored in the file "filename".

---

Detailed Description

describes structure of a 3 layer back prop neural net

Definition at line 48 of file BPnnet.H.

---

Constructor & Destructor Documentation

BPnnet::BPnnet	(	const int	numInput,
		const int	numHidden,
		const KnowledgeBase *	kb
	)

Constructor.

Definition at line 50 of file BPnnet.C.

References KnowledgeBase::getSize(), and Image< T >::resize().

BPnnet::~BPnnet

(

)

Destructor.

Definition at line 77 of file BPnnet.C.

References Image< T >::freeMem().

---

Member Function Documentation

bool BPnnet::load

(

const char *

filename

)

Assign all weights based on data stored in the file "filename".

returns false if the # of units in each layer do not match the matrix sizes in the file

Definition at line 252 of file BPnnet.C.

References Image< T >::getArrayPtr(), and Image< T >::getSize().

void BPnnet::normalizeWeights

(

void

)

Normalize weights.

Definition at line 98 of file BPnnet.C.

References getMinMax(), and inplaceClamp().

Referenced by randomizeWeights().

void BPnnet::randomizeWeights

(

void

)

Initialize all weights to (small) random values.

Definition at line 84 of file BPnnet.C.

References normalizeWeights(), randomDouble(), and Image< T >::setVal().

bool BPnnet::recognize	(	const Image< float > &	in,
		SimpleVisualObject &	vo
	)

Attempt to recognize given jet as a certain visual object.

Definition at line 196 of file BPnnet.C.

References SimpleVisualObject::getName(), KnowledgeBase::getSimpleVisualObject(), and max().

bool BPnnet::save

(

const char *

filename

)

const

Store 2 matrices of weights to the file "filename".

use this after training a net

Definition at line 232 of file BPnnet.C.

References Image< T >::getArrayPtr(), and Image< T >::getSize().

double BPnnet::train	(	const Image< float > &	in,
		const SimpleVisualObject &	target,
		const double	learnRate
	)

Do one training iteration.

returns rms1 = sum of (target output - actual output)^2/n_outputs for all output layer neurons

Definition at line 111 of file BPnnet.C.

References KnowledgeBase::findSimpleVisualObjectIndex(), SimpleVisualObject::getName(), Image< T >::getVal(), and Image< T >::setVal().

---

The documentation for this class was generated from the following files:

• src/BPnnet/BPnnet.H
• src/BPnnet/BPnnet.C