BIE::TemperedDifferentialEvolution Class Reference

Implements the Differential Evolution (DE) algorithm (see DifferentialEvolution) including Tempered Transistion steps every tfreq steps (see TemperedSimulation). More...

#include <TemperedDifferentialEvolution.h>

Inheritance diagram for BIE::TemperedDifferentialEvolution:

[legend]

Collaboration diagram for BIE::TemperedDifferentialEvolution:

[legend]

List of all members.

Public Member Functions

TemperedDifferentialEvolution (int max, int ndim, int number, string deinit, int minmc_p, double maxT, BaseDataTree *d, Model *m, Integration *i, Converge *c, MixturePrior *mp, LikelihoodComputation *l, MCAlgorithm *mca, Simulation *last=0)

Constructor for top level invocation.

void SetTempFreq (int n)

Change tempering frequency for mixing (default: 11).

void SetEquilSteps (int n)

Set number of DE steps at each temperature level (default: 10).

void SetTpow (double x)

Set the scale factor $T^{tpow}$ for proposal function.

void SetMinMC (int n)

Set the minimum number of temperature states.

virtual void PrintStepDiagnostic ()

override of Simulation method to provide diagnostics specific to TemperedDifferentialEvolution

Static Public Attributes

static int Ninter

Number of mixing steps between tempering (default 10).

static double tpow

Scale factor $T^{tpow}$ for proposal function.

static int minmc

Minimum number of temperature states.

static unsigned tfreq

Frequency in steps for a tempered step (default: 11).

Protected Member Functions

virtual void MCMethod ()

Run the entire Monte Carlo algorithm (including tempering).

virtual void OneDifferentialEvolutionStep ()

Take a single DE step only.

virtual void PowerUP (int)

Heat chains to temperature level k.

virtual void PowerDOWN (int)

Cool chains to temperature level k.

Protected Attributes

vector< double > up

The log prob of the state moving UP the temperature ladder.

vector< double > down

The log prob of the state moving DOWN the temperature ladder.

vector< double > beta

The 1/KT values for the temperature ladder.

vector< double > factor

The multiplicative width factors for the ladder.

int curswp

The number of swapped states on the current step.

int curtry

The number of swap attempts on the current step.

int totswp

The cumulative number of swapped states (since last query).

int tottry

The cmulative number of swap attempts (since last query).

double MaxT

The maximum tempering temperature.

vector< unsigned long > mstat0

Tally of proposals for each temperature level.

vector< unsigned long > stat0T

vector< unsigned long > mstat1

Tally of accepted proposals for each level.

vector< unsigned long > stat1T

bool tstat

Tempered statistics counter active.

Friends

class boost::serialization::access

Detailed Description

Implements the Differential Evolution (DE) algorithm (see DifferentialEvolution) including Tempered Transistion steps every tfreq steps (see TemperedSimulation).

Temperature levels

The temperature levels $j=0,\ldots,N$ are defined by:

$P_j \propto P_0^{j/N} = e^{\beta_j \log P_0}$

or

$P_j \propto e^{\beta_j \log P_0} \beta_j = exp(-\log(T_{max})*j/N)$

and the Metropolis-Hastings width is multiplied by the factor

$f_j = \beta_j^{-\tau}$

where $\tau$ is the paramater tpow.

Paremeters

Parameters:

	Ninter	is the number of mixing steps between tempering steps (default: 10)
	tfreq	is the number steps between tempered transition states (default: 11)
	tpow	is the scale factor $T^{tpow}$ for proposal function (default: 0.5, exact for a Gaussian)
	minmc	is the minimum number of temperature levels. If not overriddednm explicitly, the constructor will set number of levels so that there are the square root of the parameter rank per logarithmic interval in temperature. (default: 4)

Constructor & Destructor Documentation

BIE::TemperedDifferentialEvolution::TemperedDifferentialEvolution	(	int	max,
		int	ndim,
		int	number,
		string	deinit,
		int	minmc_p,
		double	maxT,
		BaseDataTree *	d,
		Model *	m,
		Integration *	i,
		Converge *	c,
		MixturePrior *	mp,
		LikelihoodComputation *	l,
		MCAlgorithm *	mca,
		Simulation *	last = `0`
	)

Constructor for top level invocation.

Initial constuctor Multiple-level constuctor

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

/home/weinberg/src/BIE/include/TemperedDifferentialEvolution.h