A Topographic Self Organizing map fp, with continuous transformation learning. More...

Inheritance diagram for CTSOMFeaturePlane:

Public Member Functions
	~CTSOMFeaturePlane ()
void	read (QDataStream inStream, QProgressDialog progress, int *index)
void	save (QDataStream outStream, QProgressDialog progress, int *index)
bool	isLearning ()
FeaturePlaneInput *	addFilter (Filter f, FeaturePlaneAbstract fp, int type, int fpOutputIndex, bool wta)
void	incWinCount ()
int	getWinCount ()
Static Public Member Functions
static void	compete (vector< FeaturePlaneAbstract * > _featurePlanes, Layer *theLayer, int step)
static void	competeOld (vector< FeaturePlaneAbstract * > _featurePlanes, Layer *theLayer, int step)
Protected Member Functions
	CTSOMFeaturePlane (Network net, int l, QString name, int w, int h, int angles, int speeds, int alpha, int speed, bool isWTA, bool visible, Neuron neuron, NotifyStrategyAbstract notify, bool learning)
void	step ()
float	nonlinearity (Matrix totalActivation, OutputImage workOutput)
void	computeGatingUnits ()
Friends
class	GatingUnits
class	FeaturePlaneFactory

Detailed Description

A Topographic Self Organizing map fp, with continuous transformation learning.

Constructor & Destructor Documentation

CTSOMFeaturePlane::~CTSOMFeaturePlane ( )

CTSOMFeaturePlane::CTSOMFeaturePlane	(	Network *	net,
		int	l,
		QString *	name,
		int	w,
		int	h,
		int	angles,
		int	speeds,
		int	alpha,
		int	speed,
		bool	isWTA,
		bool	visible,
		Neuron *	neuron,
		NotifyStrategyAbstract *	notify,
		bool	learning
	)		`[protected]`

Constructor

Parameters:

net	- network, this FP will be added to the end of the list of feature planes
l	- layer index
name	- name of the feature plane
w	- width
h	- height
angles	- number of discrete angles (e.g. speeds, orientations)
speeds	- number of speeds (e.g. speed bands, scale bands)
alpha	- alpha
speed	- speed of this fp
isWTA	- participates in Selective Tuning
visible	- should the FP be presented in the UI
neuron	- neuron strategy (what is the output nonlinearity, or the ODE that defines response)
notify	- notification strategy (deprecated, on the way out)
learning	- are weights adjustable through learning?

FeaturePlaneInput * CTSOMFeaturePlane::addFilter	(	Filter *	f,
		FeaturePlaneAbstract *	fp,
		int	type,
		int	fpOutputIndex,
		bool	wta
	)		`[virtual]`

Add an input filter/feature plane pair.

Parameters:

f	- filter
fp	- input feature plane
type	- input type: regular, non-Fourier, masking
fpOutputIndex	- in case input feature plane has more than one output, which one are we interested in

void CTSOMFeaturePlane::compete	(	vector< FeaturePlaneAbstract * >	_featurePlanes,
		Layer *	theLayer,
		int	step
	)		`[static]`

void CTSOMFeaturePlane::competeOld	(	vector< FeaturePlaneAbstract * >	_featurePlanes,
		Layer *	theLayer,
		int	step
	)		`[static]`

void CTSOMFeaturePlane::computeGatingUnits ( ) [protected]

int CTSOMFeaturePlane::getWinCount ( ) [inline]

void CTSOMFeaturePlane::incWinCount ( ) [inline]

bool CTSOMFeaturePlane::isLearning ( )

float CTSOMFeaturePlane::nonlinearity	(	Matrix *	totalActivation,
		OutputImage *	workOutput
	)		`[protected]`

void CTSOMFeaturePlane::read	(	QDataStream *	inStream,
		QProgressDialog *	progress,
		int *	index
	)		`[virtual]`

void CTSOMFeaturePlane::save	(	QDataStream *	outStream,
		QProgressDialog *	progress,
		int *	index
	)		`[virtual]`

void CTSOMFeaturePlane::step ( ) [protected, virtual]

Do one computation step - compute activations, and FP output

Reimplemented from FeaturePlane.

friend class FeaturePlaneFactory [friend]

friend class GatingUnits [friend]

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