mcmc: MCMC.h File Reference

Go to the source code of this file.


Namespaces
namespace	MCMC
Classes
class	ParameterBase
	Pure virtual class provide interface for Parameter. More...
class	ParameterT< T >
	Base class for model parameters. More...
class	StepBase
	Base class for steps associated with different parameter types. More...
class	Step< T >
	Base class for all steps associated with one parameter type. More...
class	MetroStepT< T >
	Implements Metropolis-Hastings step for a parameter. More...
class	AdaptMetroStepT< T >
	Implements Metropolis-Hastings step for a parameter. More...
class	SliceStep
	Implements a slice sampler. More...
class	FunctionStepT< T >
	Implements a deterministic node. More...
class	Model
	Implements the statistical model. More...
Defines
#define	__MCMC_H
Typedefs
typedef std::vector< StepBase * >	ModelSteps
	pointer to each Step in the Model
typedef std::vector< boost::any >	SampleVector
	parameters at one iteration
typedef SampleVector::const_iterator	SampleIter
	paramter iterator
typedef std::vector< SampleVector >	Results
	SampleVectors for all iterations.
typedef Results::const_iterator	ResultsIter
	iterator over iterations
typedef ParameterT< double >	Parameter
typedef FunctionStepT< double >	FunctionStep
typedef MetroStepT< double >	MetroStep
Functions
lot &	GetRNG ()
double	safeFreq (double p, double zero=MCMC_ZERO_FREQ)
double	invSafeFreq (double x, double zero=MCMC_ZERO_FREQ)
std::vector< double >	safeFreq (const std::vector< double > &p, const double zero)
std::vector< double >	invSafeFreq (const std::vector< double > &x, const double zero=MCMC_ZERO_FREQ)
double	proposeBeta (double mean, double denom, double tolerance)
double	logQBeta (double newMean, double oldMean, double denom, double tolerance)
std::vector< double >	proposeDirch (const std::vector< double > &mean, double denom, double tolerance)
double	logQDirch (const std::vector< double > &newMean, const std::vector< double > &oldMean, double denom, double tolerance)
double	proposeNorm (double mean, double variance)
double	logQNorm (double newMean, double oldMean, double variance)
double	safeBetaPar (double x)
void	safeDirchPar (std::vector< double > &x)
Variables
const double	MCMC_ZERO_FREQ
const double	MCMC::MinBetaPar = 1.0e-1
	minimum value allowed for beta parameter
const double	MCMC::MaxBetaPar = 1.0e4
	maximum value allowed for beta parameter
const double	MCMC::MinDirchPar = 1.0e-1
	minimum value allowed for Dirichlet parameter
const double	MCMC::MaxDirchPar = 1.0e4
	maximum value allowed for Dirichlet parameter

Detailed Description

The ParameterT, StepT, and Model classes defined here are the core classes that do all of the work. To implement a Bayesian model using these classes each parameter in the model should be derived from Paramater and should override llike(), at a minimum. lprior() should be overridden for anything other than a flat prior. (Notice that the prior will be improper unless overriden when the parameter has an unbounded domain.) The model is derived from Model, and each parameter is pushed onto a stack (step_), with a specified Step type (MetroStep, SliceStep, or FunctionStep).

Typedef Documentation

typedef FunctionStepT<double> FunctionStep

typedef for convenienc access to double paramters

Definition at line 671 of file MCMC.h.

typedef MetroStepT<double> MetroStep

typedef for convenienc access to double paramters

Definition at line 674 of file MCMC.h.

typedef ParameterT<double> Parameter

typedef for convenienc access to double paramters

Definition at line 668 of file MCMC.h.

Function Documentation

lot& GetRNG ( void )

Returns a reference to the internal random number generator

Definition at line 104 of file MCMC.cpp.

std::vector<double> invSafeFreq	(	const std::vector< double > &	x,
		const double	zero
	)

Inverts a safeFreq()ed vector

Parameters:

	x	The vector to invert
	zero	The original guard (not checked)

Definition at line 153 of file MCMC.cpp.

double invSafeFreq	(	const double	x,
		const double	zero
	)

Converts a safeFreq()ed x back

Parameters:

	x	Frequency to convert back
	zero	Value used in original guard (not verified)

Definition at line 124 of file MCMC.cpp.

Referenced by logQBeta(), and logQDirch().

double logQBeta	(	const double	newMean,
		const double	oldMean,
		const double	denom,
		const double	tolerance
	)

lQ for a beta proposal

Parameters:

	newMean	New value proposed
	oldMean	Old value from which proposed
	denom	Effective sample size
	tolerance	Minimum value allowed in original proposal

Definition at line 196 of file MCMC.cpp.

References Density::dbeta(), invSafeFreq(), safeBetaPar(), and Util::safeLog().

double logQDirch	(	const std::vector< double > &	newMean,
		const std::vector< double > &	oldMean,
		const double	denom,
		const double	tolerance
	)

lQ for a Dirichlet proposal

Parameters:

	newMean	New value proposed
	oldMean	Old value from which proposed
	denom	Effective sample size
	tolerance	Minimum value allowed in original proposal

Definition at line 242 of file MCMC.cpp.

References Density::ddirch(), invSafeFreq(), and Util::safeLog().

double logQNorm	(	const double	newMean,
		const double	oldMean,
		const double	variance
	)

lQ for a normal proposal

Parameters:

	newMean	New value proposed
	oldMean	Old value from which proposed
	variance	Variance of the proposal distribution

Definition at line 271 of file MCMC.cpp.

References Density::dnorm(), and Util::safeLog().

double proposeBeta	(	const double	mean,
		const double	denom,
		const double	tolerance
	)

Propose a new beta variate for an M-H step

Parameters:

	mean	Mean for beta distribution
	denom	Effective sample size
	tolerance	Minimum value allowed (to avoid exact 0s and 1s)

The parameters of the beta distribution are given by

$\alpha = denom \times mean$

$\beta = denom \times (1-mean)$

where alpha and beta are guarded by safeBetaPar()

Definition at line 178 of file MCMC.cpp.

References lot::beta(), safeBetaPar(), and safeFreq().

std::vector<double> proposeDirch	(	const std::vector< double > &	q,
		const double	denom,
		const double	tolerance
	)

Propose a new Dirichlet vector for an M-H step

Parameters:

	q	Mean of the proposal distribution
	denom	Effective sample size
	tolerance	Minimum value allowed (to avoid exact 0s and 1s)

The parameters of the beta distribution are given by

$\alpha = denom \times mean$

$\beta = denom \times (1-mean)$

where alpha and beta are guarded by safeBetaPar()

Definition at line 220 of file MCMC.cpp.

References lot::dirichlet(), safeDirchPar(), and safeFreq().

double proposeNorm	(	const double	mean,
		const double	variance
	)

Propose a new normal variate for an M-H step

Parameters:

	mean	Mean of the proposal distribution
	variance	Variance of the proposal distribution

Definition at line 261 of file MCMC.cpp.

References lot::norm().

double safeBetaPar ( const double x )

Ensures MinBetaPar <= x <= MaxBetaPar

Parameters:

Frequency to guard

Definition at line 282 of file MCMC.cpp.

References MCMC::MaxBetaPar, and MCMC::MinBetaPar.

Referenced by logQBeta(), and proposeBeta().

void safeDirchPar ( std::vector< double > & x )

Ensures MinBetaPar <= x <= MaxBetaPar for all elements of a Dirichlet parameter vector

Parameters:

Frequency to guard

Definition at line 293 of file MCMC.cpp.

References MCMC::MaxDirchPar, and MCMC::MinDirchPar.

Referenced by proposeDirch().

std::vector<double> safeFreq	(	const std::vector< double > &	p,
		const double	zero
	)

Ensures that all values in vector are greater than zero and less than 1.0 - n*zero, where n is p.size()

Parameters:

	p	Frequency vector to guard
	zero	Value used for guard (MCMC_ZERO_FREQ default)

Assumes all elements of p are in [0, 1]

Definition at line 138 of file MCMC.cpp.

double safeFreq	(	const double	p,
		const double	zero
	)

Ensures zero < p < 1-zero

Parameters:

	p	Frequency to guard
	zero	Minimum value of p allowed

Assumes all elements of p are in [0, 1]

Definition at line 115 of file MCMC.cpp.

Referenced by proposeBeta(), and proposeDirch().

MCMC.h File Reference

Namespaces

Classes

Defines

Typedefs

Functions

Variables

Detailed Description

Typedef Documentation

Function Documentation

Variable Documentation