ipsdk::math::clustering::SpectralClustering< KernelType > Class Template Reference

Inheritance diagram for ipsdk::math::clustering::SpectralClustering< KernelType >:

Public Types
typedef KernelKMeansClusteringConfig	AlgorithmConfig
	algorithm configuration
typedef KernelKMeansClustering< KernelType >	ClusteringAlgorithm
	clustering algorithm type
typedef boost::shared_ptr< ClusteringAlgorithm >	ClusteringAlgorithmPtr
	shared pointer to clustering algorithm

Public Member Functions
const KernelType &	getKernel () const
	retrieve kernel associated to object
void	clear ()
	clear of processing results
Constructors and destructor
	SpectralClustering (const KernelType &kernel)
	~SpectralClustering ()
const AlgorithmConfig &	getConfig () const
	access to algorithm configuration
void	setConfig (const AlgorithmConfig &config)
	access to algorithm configuration
Algorithm computation
bool	isClusteringDone () const
	check whether a computation has been done
bool	compute (const ipUInt32 nbClusters, const VectorColl &inputDataColl)
	proceed to data clustering More...
Algorithm results
ipUInt32	getNbClusters () const
	retrieve number of clusters associated to clustering process More...
ipUInt32	getDataSize () const
	retrieve data size associated to input data collection More...
ipUInt32	getNbInputData () const
	retrieve number of data in input data collection More...
const VectorColl &	getInputDataColl () const
	retrieve used input data collection More...
ipUInt32	getNbFailedAttempts () const
	retrieve number of failed attempts More...
ipReal64	getSeparation () const
	get separation associated to clustering result More...
ipUInt32	getNbIteration () const
	retrieve number of iterations associated to selected clustering solution More...
ipUInt32	getClusterPopulation (const ipUInt32 clusterIdx) const
	retrieve number of data associated to a given cluster More...
const UInt32Vector &	getAssignments () const
	retrieve cluster index associated to each input data More...
const Vector &	getEigenValues () const
	retrieve eigen values associated to spectral clustering More...
const Matrix &	getEigenVectors () const
	retrieve eigen vectors associated to spectral clustering More...
ipUInt32	computeAssignment (const Vector &data) const
	function allowing to compute cluster assignment for a given data More...
const ClusteringAlgorithm &	getClusteringAlgorithm () const
	retrieve underlying used clustering algorithm

Protected Attributes
ClusteringAlgorithmPtr	_pClusteringAlgorithm
	underlying clustering algorithm
VectorColl	_inputDataColl
	input data collection
UInt32Vector	_assignments
	cluster index associated to each input data
Vector	_eigenValues
	eigen values associated to eigen problem resolution
Matrix	_eigenVectors
	eigen vectors associated to eigen problem resolution

Member Function Documentation

compute()

template<typename KernelType>

bool ipsdk::math::clustering::SpectralClustering< KernelType >::compute	(	const ipUInt32	nbClusters,
		const VectorColl &	inputDataColl
	)

proceed to data clustering

getNbAttempts() of K-Means++ algorithm will be used to initialize K-Means algorithm.

For a given attempt, if process reach maximum number of iterations (getConfig().getNbMaxIteration()) before that clusters centers stabilization is reached, current attempt is considered as failed

If none of the getNbAttempts() succeed, algorithm is considered on failure

On output clusters are sorted in decreasing population order

Parameters

nbClusters	number of searched clusters
inputDataColl	collection of observations

Returns

false if no attempt succeed

Exceptions

ipsdk::math::IPSDKMathException	if nbClusters == 0
ipsdk::math::IPSDKMathException	if nbClusters > inputDataColl.size()

Warning

there is no check to ensure that inputDataColl[i].size() == inputDataColl[0].size()

getNbClusters()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getNbClusters

(

)

const

retrieve number of clusters associated to clustering process

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getDataSize()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getDataSize

(

)

const

retrieve data size associated to input data collection

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getNbInputData()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getNbInputData

(

)

const

retrieve number of data in input data collection

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getInputDataColl()

template<typename KernelType>

const VectorColl& ipsdk::math::clustering::SpectralClustering< KernelType >::getInputDataColl

(

)

const

retrieve used input data collection

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getNbFailedAttempts()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getNbFailedAttempts

(

)

const

retrieve number of failed attempts

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getSeparation()

template<typename KernelType>

ipReal64 ipsdk::math::clustering::SpectralClustering< KernelType >::getSeparation

(

)

const

get separation associated to clustering result

separation is defined as the Dunn validity index

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getNbIteration()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getNbIteration

(

)

const

retrieve number of iterations associated to selected clustering solution

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getClusterPopulation()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::getClusterPopulation

(

const ipUInt32

clusterIdx

)

const

retrieve number of data associated to a given cluster

Exceptions

ipsdk::math::IPSDKMathException

if clusterIdx >= getNbClusters()

getAssignments()

template<typename KernelType>

const UInt32Vector& ipsdk::math::clustering::SpectralClustering< KernelType >::getAssignments

(

)

const

retrieve cluster index associated to each input data

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getEigenValues()

template<typename KernelType>

const Vector& ipsdk::math::clustering::SpectralClustering< KernelType >::getEigenValues

(

)

const

retrieve eigen values associated to spectral clustering

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

getEigenVectors()

template<typename KernelType>

const Matrix& ipsdk::math::clustering::SpectralClustering< KernelType >::getEigenVectors

(

)

const

retrieve eigen vectors associated to spectral clustering

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

computeAssignment()

template<typename KernelType>

ipUInt32 ipsdk::math::clustering::SpectralClustering< KernelType >::computeAssignment

(

const Vector &

data

)

const

function allowing to compute cluster assignment for a given data

Exceptions

ipsdk::math::IPSDKMathException

if isClusteringDone() == false

Warning

there is no check to ensure that data.size() == getDataSize()

this assignment computation uses Nystrom approximation (see Sampling with Minimum Sum of Squared Similarities for Nystrom-Based Large Scale Spectral Clustering, Djallel Bouneffouf and Inanc Biro, Proceedings of the Twenty-Fourth IJCAI, International Joint Conference on Artificial Intelligence, 2015)

---

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

• /media/reactivipdev/Nightly/IPSDK_Patch/IPSDK_4_1/IPSDK_Nightly/include/public/IPSDKMath/Clustering/ClusteringTypes.h
• /media/reactivipdev/Nightly/IPSDK_Patch/IPSDK_4_1/IPSDK_Nightly/include/public/IPSDKMath/Clustering/SpectralClustering.h