BIE::Ensemble Class Reference

Ensemble caches an group of posterior states in a vector and provides member functions to characterize the posterior from this distribution and produce statistical diagnostics on the convergence of the simulation. More...

#include <Ensemble.h>

Inheritance diagram for BIE::Ensemble:

[legend]

Collaboration diagram for BIE::Ensemble:

[legend]

List of all members.

int	count
	State variables and vectors.
int	offset
	Offset of current count list from zero.
int	burnIn
	Offset at convergence.
int	cur_index
	The current index in the cache.
int	ibeg
	Currently unused (remove?).
int	clev
	Level on creation.
vector< double >	ccum
	Fraction of states in each subspace.
vector< pair< double, int > >	fraction
	For sampling.
deque< StateData >	states
	State cache.
Uniform *	unit
	Uniform random number generator.
DiscreteUniform *	disc
	Discrete uniform random number generator.
bool	dimensions_set
	True if the ensemble has had its mixture cardinality and model dimension set somehow (either in the constructor or by the member function).
bool	verbose
	Verbose output.
int	_id
	Local instance number.
static int	instance
	Count instances.
static int	keep
	Global variables.
static double	thresh
	Principal component threshold (default=1.0e-12).
static bool	continuous
	For variable component mixtures, specify whether to determine the range of continuously adjacent non-zero components (continuous=true, default) or the minimum and maximum (allowing for the possibility of zero frequency component counts (continuous=false).
static int	key_pos
	Which element of parameter vector to be used as index key (default=0).
int	sampleSubspace ()
	Sample from marginalized component number distribution.
void	makeSampleFraction (vector< int > &)
	Make fraction map for sampling.
ComputeDistribution
Computes the covariance matrix and its eigenvalues/vectors Return = 1 (variance computed), Return = 0 (too few values)
class	boost::serialization::access
virtual void	ComputeDistribution ()
	Compute distribution estimate as needed.
virtual void	ComputeDistribution (int n)
virtual void	stats (int n, double &m, double &s)
	Returns the mean and variance for the posterior probabilties for the last $n$ states in the ensemble.
virtual int	Nstates ()
virtual double	PDFSubspace (int m)
virtual double	logPDF (State &p)
	Log of differential distribution function.
virtual double	logPDFMarginal (int m, int n, vector< double > &V)
	Log of differential distribution function for a component in subspace.
virtual Ensemble *	New ()
	Object factory (clone).
virtual vector< double >	Mean (int m)
	Retrieve the mean per subspace.
virtual vector< double >	StdDev (int m)
	Retrieve the std dev per subspace.
virtual vector< double >	StdDev (void)
	Retrieve the std dev.
virtual vector< double >	Mean (void)
	Retrieve the mean.
virtual vector< double >	Sample (int m)
	Sample the distribution.
virtual vector< double >	Sample ()
	Sample the distribution.
virtual vector< double >	Width (int m)
	Width.
virtual vector< double >	StdDevMarginal (int m, int n)
	Retrieve the std dev for the component marginal.
virtual vector< double >	MeanMarginal (int m, int n)
	Retrieve the mean for the component marginal.
virtual vector< double >	SampleMarginal (int m, int n)
	Sample the distribution for the component marginal.
virtual vector< double >	WidthMarginal (int m, int n)
	Width.
virtual void	PrintDiag (ostream &out)
	Print current covariance matrix (debug and diag).
virtual void	PrintDiag (string &outfile)
virtual int	MinC ()
	State with minimum number of components.
virtual int	MaxC ()
	State with maximum number of components.
virtual void	Broadcast ()
	Copy state to all nodes.
Public Member Functions
	Ensemble ()
	Constructor.
	Ensemble (int nmix, int ndim)
	Mixture constructor.
	Ensemble (int nmix, int ndim, int level, int nburn, string filename, int keypos)
	General constructor.
	Ensemble (const Ensemble &)
	Copy constructor.
	~Ensemble ()
	Destructor.
virtual void	Reset (int nmix, int ndim)
	Reinitialize the Ensemble.
virtual void	Reset (int nmix, int ndim, int level, int nburn, string filename, int keypos)
	Reinitialize the Ensemble.
virtual void	setDimensions (int nmix, int ndim)
	Set total size of arrays.
virtual void	setContinuous ()
	Set multiple component range to continuous.
virtual void	setMaxRange ()
	Set multiple component range to maximum range (allow zero freq. states).
virtual void	setNKeep (int n)
	Set the maximum number of states to keep in cache.
virtual void	Order (int n)
	Order the mixture by the given component n<0 is the weight n>=0 is the parameter vector index.
virtual void	setVerboseOn ()
	Diagnostics for intermediate computation.
virtual void	setVerboseOff ()
	Turn off intermediate computation diagnostics.
virtual bool	AccumData (double, State &)
	Adds a state to the deque.
virtual bool	AccumData (StateData &s)
	Adds a state to the deque (using StateData structure).
virtual bool	AccumulateData (double v, RecordBuffer *datapoint)
	Required member but not used here.
Protected Attributes
int	M
	Number of components in the state.
vector< int >	cnts
	The number of states used in the distribution.
vector< int >	mcnt
int	minc
	Mininum number of component states (from cnts).
int	maxc
	Mmaximum number of component states (from cnts).
int	N
	Dimension per component.

---

Detailed Description

Ensemble caches an group of posterior states in a vector and provides member functions to characterize the posterior from this distribution and produce statistical diagnostics on the convergence of the simulation.

The posterior state is characterized from the ensemble variance by a principal component analysis.

---

Constructor & Destructor Documentation

BIE::Ensemble::Ensemble	(	int	nmix,
		int	ndim
	)

Mixture constructor.

Parameters:

	nmix	is the mixture number
	ndim	is the dimension of each mixture copmonent

BIE::Ensemble::Ensemble	(	int	nmix,
		int	ndim,
		int	level,
		int	nburn,
		string	filename,
		int	keypos
	)

General constructor.

Parameters:

	nmix	is the mixture number
	ndim	is the dimension of each mixture copmonent
	level	is the update level
	nburn	is the desired number of converged states
	filename	is the logfile for ensemble statistics
	keypos	is the parameter index for computing PDF and CDF evaluations

---

Member Function Documentation

virtual void BIE::Ensemble::stats	(	int	n,
		double &	m,
		double &	s
	)			[virtual]

Returns the mean and variance for the posterior probabilties for the last $n$ states in the ensemble.

If $n$ is less the number of states in the deque, the whole deque is used. If the number of states is less than 2, zero vales are returned.

virtual double BIE::Ensemble::logPDFMarginal	(	int	m,
		int	n,
		vector< double > &	V
	)			[inline, virtual]

Log of differential distribution function for a component in subspace.

Parameters:

	m,component
	n,and	parameter vector
	V.	This method is used by ReversibleJump classes.

Reimplemented in BIE::EnsembleDisc, and BIE::EnsembleStat.

Here is the caller graph for this function:

---

Member Data Documentation

int BIE::Ensemble::count [protected]

State variables and vectors.

Total count

int BIE::Ensemble::keep [static]

Global variables.

Number of states to retain (FIFO queue, default=4000)

bool BIE::Ensemble::continuous [static]

For variable component mixtures, specify whether to determine the range of continuously adjacent non-zero components (continuous=true, default) or the minimum and maximum (allowing for the possibility of zero frequency component counts (continuous=false).

These can be set in the CLI using the SetContinuous and SetMaxRange member functions

---

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

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