stateline::mcmc Namespace Reference

Classes
class	ChainArray
	Manager for all the states and handle reading / writing from / to databse. More...
class	EpsrConvergenceCriteria
	Convergence test using estimated potential scale reduction (EPSR). It is a convergence metric that takes into account the variance of the means between chains and the variance of the samples within each chain. More...
class	Sampler
	Represents a Markov Chain Monte Carlo sampler that returns samples from a distribution. More...
struct	State
	Used to represent a state in a Markov Chain. More...

Typedefs
using	PropAcceptFn = std::function< bool(const State &, const State &, double)>
	Type representing proposal acceptance functions.
using	SwapAcceptFn = std::function< bool(const State &, const State &, double, double)>
	Type representing swap acceptance functions.

Enumerations
enum	SwapType { SwapType::NoAttempt, SwapType::Accept, SwapType::Reject }
	Used for recording swapping of chains. More...

Functions
Eigen::VectorXd	bouncyBounds (const Eigen::VectorXd &val, const Eigen::VectorXd &min, const Eigen::VectorXd &max)
	A function to bounce the MCMC proposal off the hard boundaries. This allows the proposal to always move around without getting stuck at 'walls'. More...
Eigen::VectorXd	adaptiveGaussianProposal (const Eigen::VectorXd &state, double sigma, const Eigen::VectorXd &min, const Eigen::VectorXd &max)
	An adaptive Gaussian proposal function. It randomly varies each value in the state according to a Gaussian distribution whose variance changes depending on the acceptance ratio of a chain. It also bounces of the walls of the hard boundaries given so as not to get stuck in corners. More...
bool	acceptProposal (const State &newState, const State &oldState, double beta)
	Returns true if we want to accept the MCMC step. More...
bool	acceptSwap (const State &stateLow, const State &stateHigh, double betaLow, double betaHigh)
	Returns true if we want to accept the MCMC swap. More...

Detailed Description

Namespace for MCMC sampling features.

Enumeration Type Documentation

enum stateline::mcmc::SwapType

strong

Used for recording swapping of chains.

Enumerator
NoAttempt	No swap was attempted.
Accept	A swap was accepted.
Reject	A swap was rejected.

Function Documentation

bool stateline::mcmc::acceptProposal	(	const State &	newState,
		const State &	oldState,
		double	beta
	)

Returns true if we want to accept the MCMC step.

Parameters

newState	The proposed state.
oldState	The current state of the chain.
beta	The inverse temperature of the chain.

Returns

True if the proposal was accepted.

bool stateline::mcmc::acceptSwap	(	const State &	stateLow,
		const State &	stateHigh,
		double	betaLow,
		double	betaHigh
	)

Returns true if we want to accept the MCMC swap.

Parameters

stateLow	The state of the lower temperature chain.
stateHigh	The state of the higher temperature chain.
betaLow	The inverse temperature of the lower temperature chain.
betaHigh	The inverse temperature of the high temperature chain.

Returns

True if the swap was accepted.

Eigen::VectorXd stateline::mcmc::adaptiveGaussianProposal	(	const Eigen::VectorXd &	state,
		double	sigma,
		const Eigen::VectorXd &	min,
		const Eigen::VectorXd &	max
	)

An adaptive Gaussian proposal function. It randomly varies each value in the state according to a Gaussian distribution whose variance changes depending on the acceptance ratio of a chain. It also bounces of the walls of the hard boundaries given so as not to get stuck in corners.

Parameters

state	The current state of the chain
sigma	The standard deviation of the distribution (step size of the proposal)
min	The minimum bound of theta
max	The maximum bound of theta

Returns

The new proposed theta

Eigen::VectorXd stateline::mcmc::bouncyBounds	(	const Eigen::VectorXd &	val,
		const Eigen::VectorXd &	min,
		const Eigen::VectorXd &	max
	)

A function to bounce the MCMC proposal off the hard boundaries. This allows the proposal to always move around without getting stuck at 'walls'.

Parameters

val	The proposed value
min	The minimum bound of theta
max	The maximum bound of theta

Returns

The new bounced theta definitely in the bounds