Som< T > Class Template Reference

#include <Som.h>

Public Member Functions
	Som (int width=0, int height=0)
void	setGeometry (int width, int height)
void	fill (const T &sample)
void	initialise (void(*init)(T &x))
	Initialise the map with a user-defined function.
void	setNeighbourhoodRadius (double r)
void	setLearningCoefficient (double c)
void	setDistanceFunction (double(*d)(const T &x1, const T &x2))
void	setMixingFunction (void(*mix)(const T &x1, const T &x2, double coeff, T &m))
int	getHeight () const
	The number of rows in the map.
int	getWidth () const
	The number of columns in the map.
void	findBestMatchingUnit (const T &sample, int &iBest, int &jBest)
	Finds the unit with state closesest to that of sample.
void	train (const T &x)
	Training with a single sample.
void	train (const T &x, int iBest, int jBest)
void	toStream (std::ostream &ostr) const
double	defaultNeighborhoodFunction (const int &i, const int &j) const

Static Public Member Functions
static double	defaultDistanceMetric (const T &x1, const T &x2)
static void	defaultMixingFunction (const T &x1, const T &x2, double coeff, T &m)

Public Attributes
std::vector< std::vector< T > >	map
	The actual neural network, two-dimensional map of units.

Protected Attributes
double(*	distanceFunction )(const T &x1, const T &x2)
	The similarity metric applied in finding the best-matching unit.
void(*	mixingFunction )(const T &x1, const T &x2, double alpha, T &result)
	Given objects x1 and x1 and a mixing coefficient coeff , outputs mixed object m.
std::string	title
int	width
int	height
double	radius2
	The radius of the neighbourhood kernel applied in training;.
double	learningCoefficient
	The radius of the neighbourhood kernel applied in training;.

Detailed Description

template<class T = std::vector<double>>
class drain::Som< T >

The Self-Organising Map by Teuvo Kohonen.

The template

The coordinates of the map apply matrix convention ie. map[i][j] refers to element j on row i. The map is implemented a std::vector<std::vector<T> > .

Member Function Documentation

fill()

template<class T = std::vector<double>>

void fill ( const T &  sample )

inline

Set each neuron in the map to given value.

findBestMatchingUnit()

template<class T = std::vector<double>>

void findBestMatchingUnit ( const T &  sample, int &  iBest, int &  jBest  )

inline

Finds the unit with state closesest to that of sample.

sample - the state against which the similarity of unit states is compared.

i - the row of the best matching unit

j - the column of the best matching unit

initialise()

template<class T = std::vector<double>>

void initialise ( void(*)(T &x)  init )

inline

Initialise the map with a user-defined function.

The user defined function with signature

void (* init)(T & x)

should initialize

setGeometry()

template<class T = std::vector<double>>

void setGeometry ( int  width, int  height  )

inline

Parameters

height

the height recipient set

train() [1/2]

template<class T = std::vector<double>>

void train ( const T &  x )

inline

Training with a single sample.

Calls findBestMatchingUnit() and uses *mixingFunction to tune the unit.

train() [2/2]

template<class T = std::vector<double>>

void train ( const T &  x, int  iBest, int  jBest  )

inline

Training with a single sample.

Member Data Documentation

mixingFunction

template<class T = std::vector<double>>

void(* mixingFunction) (const T &x1, const T &x2, double alpha, T &result)

protected

Given objects x1 and x1 and a mixing coefficient coeff , outputs mixed object m.

One might think of mixing two std::strings, for example.

---

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

• src/drain/util/Som.h