PAMClustering Class Reference

#include <PAMClustering.h>

Public Types
typedef ublas::symmetric_matrix < double, ublas::lower >	matrix_type
typedef std::vector< unsigned int >	ivec
typedef std::vector< double >	dvec

Public Member Functions
	PAMClustering (const matrix_type &d, unsigned int k)
	~PAMClustering ()
	PAMClustering (PAMClustering &other)
	Copy constructor (copy assignment not allowed because of const referenced distance matrix) More...
int	cluster (unsigned int maxIterations=10000)
double	calcSilhouette ()
double	silCluster (unsigned int c)
const dvec &	sils () const
const ivec &	labels () const
const ivec &	medoids () const
unsigned int	numClusters () const

Protected Member Functions
void	label ()
	Label each point with its nearest medoid. More...
bool	swap ()
bool	isSelected (unsigned int i)
double	avgDist (unsigned int i, unsigned int c)
double	label (ivec &labels)

Detailed Description

Implementation of Partitioning Around Medoids algorithm. After "Finding Groups in Data: an Introduction to Cluster Analysis" by Kaufman,Rosseeuw 1990.

Todo:

Switch from brute force swap-cost calculation to more efficient cost-update.

Member Typedef Documentation

typedef std::vector<double> PAMClustering::dvec

typedef std::vector<unsigned int> PAMClustering::ivec

typedef ublas::symmetric_matrix<double,ublas::lower> PAMClustering::matrix_type

Constructor & Destructor Documentation

PAMClustering::PAMClustering	(	const matrix_type &	d,
		unsigned int	k
	)

Constructor (does not perform clustering, call cluster() to do this).

Parameters

d	symmetric distance matrix
k	number of clusters

Remarks

Input matrix d is stored as const reference, so make sure it is valid throughout the existence of your PAMClustering instance.

References label().

PAMClustering::~PAMClustering

(

)

PAMClustering::PAMClustering

(

PAMClustering &

other

)

Copy constructor (copy assignment not allowed because of const referenced distance matrix)

Member Function Documentation

double PAMClustering::avgDist ( unsigned int  i, unsigned int  c  )

protected

Returns

average distance from point i to points in cluster c

Referenced by calcSilhouette().

double PAMClustering::calcSilhouette

(

)

Calculate per point, per cluster and overall silhouette width.

Returns

overall silhouette width

References avgDist().

Referenced by AutomaticPAM::cluster(), and EventNavigator::computeClustering().

int PAMClustering::cluster

(

unsigned int

maxIterations = 10000

)

Parameters

maxIterations

is the number of iterations after which to abort

Returns

number of iterations needed until convergence

References swap().

Referenced by AutomaticPAM::cluster(), and EventNavigator::computeClustering().

bool PAMClustering::isSelected ( unsigned int  i )

protected

Returns

true if point i already was used as medoid

Referenced by swap().

void PAMClustering::label ( )

protected

Label each point with its nearest medoid.

Referenced by PAMClustering(), and swap().

double PAMClustering::label ( ivec &  labels )

protected

const ivec& PAMClustering::labels ( ) const

inline

Returns

const reference to vector of labels

Referenced by EventNavigator::computeClustering().

const ivec& PAMClustering::medoids ( ) const

inline

Returns

const reference to vector of medoids

Referenced by EventNavigator::computeClustering().

unsigned int PAMClustering::numClusters ( ) const

inline

Referenced by AutomaticPAM::cluster().

double PAMClustering::silCluster

(

unsigned int

c

)

Returns

silhouette width of cluster c as determined by last calcSilhouette() call

const dvec& PAMClustering::sils ( ) const

inline

Returns

array of silhouette width per point as determined by last calcSilhouette() call

bool PAMClustering::swap ( )

protected

Perform PAM swap step.

Returns

true if final medoids found, false otherwise

References isSelected(), and label().

Referenced by cluster().