An abstract type to calculate Gelman-Rubin convergence diagnostics for an MCMC process. More...

Inheritance diagram for subpavings::MCMCGRAuto::Diagnostic:

Public Member Functions
virtual	~Diagnostic ()
	Destructor.
virtual bool	isRequired () const =0
	Get whether this diagnostic is required for burn-in.
virtual void	clean ()=0
	Clean old calculations out.
virtual void	initialise (size_t ch, size_t res, bool logs=false)=0
	Clean old calculations out and initialise for new calculations.
virtual void	initialiseChainValues (const std::vector< ChangeOfStateInformationAutoMCMC > &infoObjs)=0
	Initialise values for each chain.
virtual void	updateChainValuesInLoop (const std::vector< ChangeOfStateInformationAutoMCMC > &infoObjs)=0
	Update calculations for each chain.
virtual int	calcDiagnosticsForLoop ()=0
	Do convergence diagnostic calculations for a loop or iteration of the MCMC process, ie using current values assuming all chains have been iterated through for this loop.
virtual void	outputResults (const std::string &filenameGRScalars, const RealVec &sampledInd, int precData=5) const =0
	Output results including the contents of sampledIndPtr.
virtual void	outputResults (const std::string &filenameGRScalars, int precData=5) const =0
	Output results.
virtual void	outputCalculations (const std::string &filenameGRWorkingCalcs, int precData=5) const =0
	Output values used in the calculations.
virtual const std::vector < std::vector< cxsc::real > > &	getScalarsRef () const =0
	Get a const reference to the scalars held by this.
virtual cxsc::real	getCurrentRhatValue () const =0
	Get the current estimated R held.
virtual std::string	getGRWorkingCalcsFilename () const =0
	Get the name used by this in output files for working calculations.
virtual std::string	getGRDiagnosticsFilename () const =0
	Get the name used by this in results files.
virtual std::string	getScalarsName () const =0
	Get the name used by this for the scalar type it is using.
virtual cxsc::real	getTolerance () const =0
	Get the tolerance used by this.

Detailed Description

An abstract type to calculate Gelman-Rubin convergence diagnostics for an MCMC process.

Each concrete type must be able to calculate its $\widehat{R}$ (from within-chain and between chain variances) and report whether this is within a specified tolerance of 1.0. Each concrete type may also be configured as required for burn-in or not.

Required for burn-in is used by objects assessing burn-in over a variety of diagnostics. Requried for burn-in indicates that even if other diagnostics have satisfactory $\widehat{R}$ values, if this does not then an MCMCGRAuto should not deem that burn-in has been achieved.

Member Function Documentation

virtual int subpavings::MCMCGRAuto::Diagnostic::calcDiagnosticsForLoop ( ) [pure virtual]

Do convergence diagnostic calculations for a loop or iteration of the MCMC process, ie using current values assuming all chains have been iterated through for this loop.

Returns:: 1 if convergence criteria satisfied, 0 otherwise

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual cxsc::real subpavings::MCMCGRAuto::Diagnostic::getCurrentRhatValue ( ) const [pure virtual]

Get the current estimated R held.

Note:: Used for debugging. *

Returns:: The current estimated R value.

Precondition:: This holds at least one estimated R.

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::initialise	(	size_t	ch,
		size_t	res,
		bool	logs = `false`
	)		`[pure virtual]`

Clean old calculations out and initialise for new calculations.

Parameters:

ch	The number of chains to initialise for.
ch	Suggested number of loops to reserve space for: this can choose to reserve more but should not reserve less capacity than this.
logs	Indicator for whether this is to keep additional logs for calculations.

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::initialiseChainValues ( const std::vector< ChangeOfStateInformationAutoMCMC > & infoObjs ) [pure virtual]

Initialise values for each chain.

Parameters:

infoObjs The collection of ChangeOfStateInformationAutoMCMC objects that this can ask to initialise its scalar values for each chain.

Precondition:: infoObjs.size() == chains .

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::outputCalculations	(	const std::string &	filenameGRWorkingCalcs,
		int	precData = `5`
	)		const `[pure virtual]`

Output values used in the calculations.

Note:: Used for debugging. Results go to filename returned by getGRWorkingCalcsFilename().

Outputs the intermediary values held and calculations for the the Gelman-Rubin convergence diagnostic calculations.

Parameters:

precData The precision for the output.

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::outputResults	(	const std::string &	filenameGRScalars,
		const RealVec &	sampledInd,
		int	precData = `5`
	)		const `[pure virtual]`

Output results including the contents of sampledIndPtr.

Note:: Results go to filename returned by getGRDiagnosticsFilename().

Outputs the results of the Gelman-Rubin convergence diagnostic calculations (W, B, estimated variance, estimated R, whether convergence criteria satisfied, and sampledIndPtr values).

Note:: This version used for debugging.

Parameters:

sampledIndPtr	A pointer to a collection of values, each one corresponding to a state for which this holds calcuations, indicating whether sampling took place in that state.
precData	The precision for the output.

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::outputResults	(	const std::string &	filenameGRScalars,
		int	precData = `5`
	)		const `[pure virtual]`

Output results.

Note:: Results go to filename returned by getGRDiagnosticsFilename().

Outputs the results of the Gelman-Rubin convergence diagnostic calculations (W, B, estimated variance, estimated R, whether convergence criteria satisfied).

Parameters:

precData The precision for the output.

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

virtual void subpavings::MCMCGRAuto::Diagnostic::updateChainValuesInLoop ( const std::vector< ChangeOfStateInformationAutoMCMC > & infoObjs ) [pure virtual]

Update calculations for each chain.

Parameters:

infoObjs The collection of ChangeOfStateInformationAutoMCMC objects that this can ask for updates to its scalar values for each chain.

Precondition:: infoObjs.size() == chains .

Implemented in subpavings::MCMCGRDiagnosticInterval, and subpavings::MCMCGRDiagnosticPSRF.

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

MCMCGRAuto.hpp

Public Member Functions

Detailed Description

Member Function Documentation