BIE::MixturePrior Class Reference

Abstract class. More...

#include <MixturePrior.h>

Inheritance diagram for BIE::MixturePrior:

[legend]

Collaboration diagram for BIE::MixturePrior:

[legend]

List of all members.

Public Member Functions
void	setMaxIter (int i)
	Set maximum number of iterations in state sampler to generate a state within the desired bounds.
void	setMinMaxWeight (double minw, double maxw)
	Set the minimum and maximum weight values.
void	getMinMaxWeight (double &minw, double &maxw)
	Return the minimum and maximum weights.
virtual void	SamplePrior (int &, vector< double > &, vector< vector< double > > &)=0
	Sample: return weights and components separately.
virtual void	SamplePrior (int, int, vector< double > &)=0
	Sample: a parameter vector for a single component in a subspace.
virtual void	SampleProposal (Chain &ch, MHWidth *width)
	Based on prior type, return random variates for mixture model.
virtual double	BirthWeight (int M)
	Choose a new component weight conditional on old weights.
virtual double	BirthWeightPDF (int M, double p)
	Probability of new weight and old weights.
virtual int	MinC ()
	Minimum number of components.
virtual int	MaxC ()
	Maximum number of components.
virtual void	setOrdering (int icomp)
	Set ordering parameters.
virtual void	unsetOrdering ()
	Remove ordering.
virtual bool	Order ()
	Return ordering flag.
virtual int	Oindx ()
	Return parameter index for ordering.
Constructors
	MixturePrior ()
	For cloning only: makes an uninitialized prior.
	MixturePrior (MixturePrior *)
	Copy constructor for derived classes.
	MixturePrior (int nmix, string pfile)
	The useful constructor.
	MixturePrior (int nmix, double alpha, string pfile)
	The even more useful constructor.
	MixturePrior (int nmix, double alpha, double pmean, string pfile)
	Constructor for a variable number of components.
	~MixturePrior ()
	Destructor.
Distribution members
The packing order in vectors is: Nmix weights, followed by Nmix vectors of Ndim elements
virtual MixturePrior *	New ()=0
	Object factor (clone).
virtual double	PDF (State &)=0
	Differential distribution function P(x).
virtual double	logPDF (State &)=0
	Log of differential distribution function P(x).
virtual double	logPDFMarginal (int M, int n, vector< double > &V)=0
	Log of differential distribution function P(x) for subspace.
virtual vector< double >	lower (void)=0
	Lower bound on distribution.
virtual vector< double >	upper (void)=0
	Upper bound on distribution.
virtual int	SampleM (void)
	Sample the number of components.
virtual vector< double >	Sample (void)
	Return random variate for variable dimension model.
virtual vector< double >	Sample (int m)=0
	Return random variate from distribution.
Static Public Attributes
static unsigned	ITMAX
	Maximum number of tries to find a good state.
static double	min_prob
	Very small but non-zero probability "epsilon".
Protected Member Functions
double	GenGamma (double a)
	Gamma-distributed variates.
double	GenBeta (double a, double b)
	Beta-distributed variates.
Protected Attributes
bool	Ordered
	Is the mixtured ordered to break the label switching?
int	Iorder
	The index of the parameter to order.
int	Nmix
	Number of components in the mixture.
vector< double >	ret
	Return vectore.
double	Alpha
	Dirichlet distribution parameter.
vector< Dirichlet * >	mix
	One distribution for each dimension.
Uniform *	unit
	Uniform random variates in [-1.0, 1.0].
Uniform *	useg
	Uniform random variates in [0.0, 1.0].
Normal *	normal
	Normal with zero mean and unit variance.
double	pmean
	Poisson distribution mean.
Poisson *	pois
	Poisson distribution.
double	minW
	Minimum and maximum weight values (default: 0, 1).
double	maxW
Friends
class	boost::serialization::access

---

Detailed Description

Abstract class.

A prior for a mixture model.

The prior for the mixture parameters will be the Dirichlet distribution. See the Prior class for a definition of distribution types for each element of the model parameter vector.

---

Constructor & Destructor Documentation

BIE::MixturePrior::MixturePrior	(	int	nmix,
		string	pfile
	)

The useful constructor.

Parameters:

	nmix	is the number of components in the mixture
	pfile	is defintion of the prior distribution for parameter vector

BIE::MixturePrior::MixturePrior	(	int	nmix,
		double	alpha,
		string	pfile
	)

The even more useful constructor.

Parameters:

alpha

is the Dirichlet distribution parameter

BIE::MixturePrior::MixturePrior	(	int	nmix,
		double	alpha,
		double	pmean,
		string	pfile
	)

Constructor for a variable number of components.

Parameters:

pmean

is Poisson mean for the prior distribution of components

---

Member Function Documentation

virtual double BIE::MixturePrior::logPDFMarginal	(	int	M,
		int	n,
		vector< double > &	V
	)			[pure virtual]

Log of differential distribution function P(x) for subspace.

Parameters:

	M	and component
	n	for parameter vector
	V	This method is used for ReversibleJump classes.

Implemented in BIE::InitialMixturePrior, and BIE::PostMixturePrior.

virtual void BIE::MixturePrior::setOrdering

(

int

icomp

)

[inline, virtual]

Set ordering parameters.

The otherwise identical mixture components are ordered by a particular value to remove the Nmix!-fold degeneracy (the so-called labelling problem). The value icomp is rank in the parameter vector to be ordered icomp=-1 signals ordering on the weight value.

---

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

• /home/weinberg/src/BIE/include/MixturePrior.h