Expectation Maximization With Gaussian Mixture Model. More...

#include <statistics.h>

Public Member Functions
	ExpMax (size_t rank, size_t k=2)
	Constructor for k-means class. More...

void	setk (size_t ngroups)
	Update the number of groups. Note that this invalidates any current information. More...

void	updateMeanCovTau (const Ref< const MatrixXd > newmeans, const Ref< const MatrixXd > newcovs, const Ref< const VectorXd > tau)
	Sets the mean matrix. Each row of the matrix is a ND-mean, where N is the number of columns. More...

void	updateMeanCovTau (const Ref< const MatrixXd > samples, Ref< MatrixXd > prob)
	Updates the mean/cov/tau coordinates by using the weighted class estimates (rather than hard classification, like the previous) More...

double	expectation (const Ref< const MatrixXd > samples, Ref< MatrixXd > prob)
	Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes. More...

Eigen::VectorXi	classify (const Ref< const MatrixXd > samples)
	Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes. More...

size_t	classify (const Ref< const MatrixXd > samples, Ref< VectorXi > oclass)
	Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes. More...

int	update (const Ref< const MatrixXd > samples, bool reinit=false)
	Updates the classifier with new samples, if reinit is true then no prior information will be used. If reinit is false then any existing information will be left intact. In Kmeans that would mean that the means will be left at their previous state. More...

const MatrixXd &	getMeans ()
	Returns the current mean matrix. More...

const MatrixXd &	getCovs ()
	Returns the current mean matrix. More...

void	compute (const Ref< const MatrixXd > samples)
	Alias for updateClasses with reinit = true. This will perform a classification scheme on all the input samples. More...

Public Attributes
const int	ndim
	Number of dimensions, must be set at construction. This is the number of columns in input samples. More...

int	maxit
	Maximum number of iterations. Set below 0 for infinite. More...

Protected Attributes
bool	m_valid
	Whether the classifier has been initialized yet. More...

Detailed Description

Expectation Maximization With Gaussian Mixture Model.

Definition at line 833 of file statistics.h.

Constructor & Destructor Documentation

npl::ExpMax::ExpMax	(	size_t	rank,
		size_t	k = `2`
	)

Constructor for k-means class.

Parameters

rank	Number of dimensions in input samples.
k	Number of groups to classify samples into

Member Function Documentation

Eigen::VectorXi npl::ExpMax::classify ( const Ref< const MatrixXd > samples )

virtual

Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes.

Parameters

samples Set of samples, 1 per row

Returns: Vector of classes, rows match up with input sample rows

Implements npl::Classifier.

size_t npl::ExpMax::classify	(	const Ref< const MatrixXd >	samples,
		Ref< VectorXi >	oclass
	)

virtual

Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes.

Parameters

samples	Set of samples, 1 per row
oclass	Output classes. This vector will be resized to have the same number of rows as samples matrix.

Implements npl::Classifier.

void npl::Classifier::compute ( const Ref< const MatrixXd > samples )

inlineinherited

Alias for updateClasses with reinit = true. This will perform a classification scheme on all the input samples.

Parameters

samples Samples, S x D matrix with S is the number of samples and D is the dimensionality. This must match the internal dimension count.

Definition at line 713 of file statistics.h.

double npl::ExpMax::expectation	(	const Ref< const MatrixXd >	samples,
		Ref< MatrixXd >	prob
	)

Given a matrix of samples (Samples x Dims, sample on each row), apply the classifier to each sample and return a vector of the classes.

Parameters

samples	Set of samples, 1 per row
prob	Class probability of each sample, for each of the potential distributions

Returns: Change in summed likelihood

const MatrixXd& npl::ExpMax::getCovs ( )

inline

Returns the current mean matrix.

Returns: The current covariance matrix, with each covariance matrix stacked on top of the next.

Definition at line 936 of file statistics.h.

const MatrixXd& npl::ExpMax::getMeans ( )

inline

Returns the current mean matrix.

Returns: The current mean matrix

Definition at line 928 of file statistics.h.

void npl::ExpMax::setk ( size_t ngroups )

Update the number of groups. Note that this invalidates any current information.

Parameters

ngroups Number of groups to classify

int npl::ExpMax::update	(	const Ref< const MatrixXd >	samples,
		bool	reinit = `false`
	)

virtual

Updates the classifier with new samples, if reinit is true then no prior information will be used. If reinit is false then any existing information will be left intact. In Kmeans that would mean that the means will be left at their previous state.

Parameters

samples	Samples, S x D matrix with S is the number of samples and D is the dimensionality. This must match the internal dimension count.
reinit	whether to reinitialize the classifier before updating

return 0 if converged, -1 otherwise

Implements npl::Classifier.

void npl::ExpMax::updateMeanCovTau	(	const Ref< const MatrixXd >	newmeans,
		const Ref< const MatrixXd >	newcovs,
		const Ref< const VectorXd >	tau
	)

Sets the mean matrix. Each row of the matrix is a ND-mean, where N is the number of columns.

Parameters

newmeans	Matrix with new mean, means are stacked so that each row represents a group mean
newmeans	Matrix with new coviance matrices. Covariance matrices are stacked so that row ndim*k gets the first element of the k'th covance matrix.
newcovs	the new covariance matrices to set in the classifier
tau	the prior probaibilities of each of the mixture gaussians

void npl::ExpMax::updateMeanCovTau	(	const Ref< const MatrixXd >	samples,
		Ref< MatrixXd >	prob
	)

Updates the mean/cov/tau coordinates by using the weighted class estimates (rather than hard classification, like the previous)

Parameters

samples	Matrix of samples, where each row is an ND-sample.
prob	Probability that each sample is in a given distribution

Member Data Documentation

bool npl::Classifier::m_valid

protectedinherited

Whether the classifier has been initialized yet.

Definition at line 732 of file statistics.h.

int npl::Classifier::maxit

inherited

Maximum number of iterations. Set below 0 for infinite.

Definition at line 727 of file statistics.h.

const int npl::Classifier::ndim

inherited

Number of dimensions, must be set at construction. This is the number of columns in input samples.

Definition at line 716 of file statistics.h.

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

/home/micahc/coding/npl/lib/statistics.h

Public Member Functions

Public Attributes

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation