SOM Class Reference

self-organised map class. More...

#include <som.h>

Inherits AbstractModel.

Inheritance diagram for SOM:

[legend]Collaboration diagram for SOM:

[legend]List of all members.

Public Types
typedef std::list< std::pair< int, double > >	Neighbours
typedef std::vector< std::pair< matrix::Matrix, double > >	Neighbourhood
Public Member Functions
	SOM ()
	SOM (unsigned int dimensions, double sigma, double eps, double rbfsize, const std::string &name="SOM", const std::string &revision="$Id: som.h,v 1.5 2008/04/17 14:54:45 martius Exp $")
	create a som
virtual	~SOM ()
virtual void	init (unsigned int inputDim, unsigned int outputDim, double unit_map=0.0, RandGen *randGen=0)
	initialised som
virtual const matrix::Matrix	process (const matrix::Matrix &input)
	passive processing of the input (this function is not constant since a recurrent network for example might change internal states
virtual const matrix::Matrix	learn (const matrix::Matrix &input, const matrix::Matrix &nom_output, double learnRateFactor=1)
virtual void	damp (double damping)
	damps the weights and the biases by multiplying (1-damping)
virtual unsigned int	getInputDim () const
	returns the number of input neurons
virtual unsigned int	getOutputDim () const
	returns the number of output neurons
virtual bool	store (FILE *f) const
	stores the object to the given file stream (binary).
virtual bool	restore (FILE *f)
	loads the object from the given file stream (binary).
virtual void	printWeights (FILE *f) const
const Neighbourhood &	getNeighbourhood ()
Public Attributes
double	eps
	learning rate for weight update
Protected Member Functions
void	initNeighbourhood (double sigma)
	initialised neighbourhood
Neighbours	getNeighbours (int winner)
	returns neighbourhood as a list of indices with weights
Static Protected Member Functions
static double	activationfunction (void *rdfsize, double d)
	activation function (rbf)
static bool	validCoord (const matrix::Matrix &m, int size)
	checks whether the given coordinate is within the lattice
static int	coordToIndex (const matrix::Matrix &m, int size)
	converts index to coordinates (size is the size of the space)
static matrix::Matrix	indexToCoord (int index, int size, int dimensions)
	converts coordinates to index (size is the size of the space)

---

Detailed Description

self-organised map class.

Implementation follows normal vector-quantiser scheme. The output of the network is for each neuron.

---

Member Typedef Documentation

typedef std::vector<std::pair<matrix::Matrix,double> > Neighbourhood

typedef std::list< std::pair<int,double> > Neighbours

---

Constructor & Destructor Documentation

SOM

(

)

SOM	(	unsigned int	dimensions,
		double	sigma,
		double	eps,
		double	rbfsize,
		const std::string &	name = "SOM",
		const std::string &	revision = "$Id: som.h,v 1.5 2008/04/17 14:54:45 martius Exp $"
	)

create a som

Parameters:

dimensions

number of dimensions of the neuron lattice

virtual ~SOM

(

)

[inline, virtual]

---

Member Function Documentation

double activationfunction	(	void *	rdfsize,
		double	d
	)			[static, protected]

activation function (rbf)

int coordToIndex	(	const matrix::Matrix &	m,
		int	size
	)			[static, protected]

converts index to coordinates (size is the size of the space)

virtual void damp

(

double

damping

)

[inline, virtual]

damps the weights and the biases by multiplying (1-damping)

Implements AbstractModel.

virtual unsigned int getInputDim

(

)

const [inline, virtual]

returns the number of input neurons

Implements AbstractModel.

const Neighbourhood& getNeighbourhood

(

)

[inline]

SOM::Neighbours getNeighbours

(

int

winner

)

[protected]

returns neighbourhood as a list of indices with weights

virtual unsigned int getOutputDim

(

)

const [inline, virtual]

returns the number of output neurons

Implements AbstractModel.

Matrix indexToCoord	(	int	index,
		int	size,
		int	dimensions
	)			[static, protected]

converts coordinates to index (size is the size of the space)

void init	(	unsigned int	inputDim,
		unsigned int	outputDim,
		double	unit_map = 0.0,
		RandGen *	randGen = 0
	)			[virtual]

initialised som

Parameters:

	inputDim	dimension of input vector
	outputDim	number of outputneurons (must be a multiple of "dimensions" given at constructor)
	unit_map	if zero then weights are randomly choosed, otherwise uniformly distributed in the inputspace of size (unit_map x unit_map x ...)

initialise neighbourhood

Implements AbstractModel.

void initNeighbourhood

(

double

sigma

)

[protected]

initialised neighbourhood

normalised gaussian

const Matrix learn	(	const matrix::Matrix &	input,
		const matrix::Matrix &	nom_output,
		double	learnRateFactor = 1
	)			[virtual]

Implements AbstractModel.

void printWeights

(

FILE *

f

)

const [virtual]

const Matrix process

(

const matrix::Matrix &

input

)

[virtual]

passive processing of the input (this function is not constant since a recurrent network for example might change internal states

Implements AbstractModel.

bool restore

(

FILE *

f

)

[virtual]

loads the object from the given file stream (binary).

Implements Storeable.

bool store

(

FILE *

f

)

const [virtual]

stores the object to the given file stream (binary).

Implements Storeable.

bool validCoord	(	const matrix::Matrix &	m,
		int	size
	)			[static, protected]

checks whether the given coordinate is within the lattice

---

Member Data Documentation

double eps

learning rate for weight update

---

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

• som.h
• som.cpp

---