#include <sample_adaptive_mh.h>

Public Types
typedef Base_step::Target_distribution	Target_distribution

typedef Mv_gaussian_proposer < SimpleVector >	Proposer

Public Member Functions
	Simple_adaptive_mh_step (const Target_distribution &target, const kjb::Matrix &initial_covariance, double goal_accept_rate)

void	set_temperature (double t)

void	set_inverse_learning_rate (double C, double alpha)

void	set_constant_learning_rate (double gamma, size_t change_point=0)

Step_log< SimpleVector >	operator() (SimpleVector &in, double lt_m)

const kjb::Matrix &	get_cholesky_covariance () const

double	get_global_scale () const

virtual void	adapt (double accept_prob, const SimpleVector &previous_state, const SimpleVector &proposed_state, const SimpleVector &accepted_state)

void	set_post_callback (const boost::function1< void, const Self & > &cb)

double	get_current_gamma () const
	returns the current learning rate (for debugging) More...

double	get_log_lambda () const
	returns the log of the current scaling lambda More...

Protected Member Functions
void	set_target (const Target_distribution &t)

Member Typedef Documentation

template<class SimpleVector>

typedef Mv_gaussian_proposer<SimpleVector> Simple_adaptive_mh_step< SimpleVector >::Proposer

template<class SimpleVector>

typedef Base_step::Target_distribution Simple_adaptive_mh_step< SimpleVector >::Target_distribution

Constructor & Destructor Documentation

template<class SimpleVector>

Simple_adaptive_mh_step< SimpleVector >::Simple_adaptive_mh_step	(	const Target_distribution &	target,
		const kjb::Matrix &	initial_covariance,
		double	goal_accept_rate
	)

inline

Member Function Documentation

template<class SimpleVector>

virtual void Simple_adaptive_mh_step< SimpleVector >::adapt	(	double	accept_prob,
		const SimpleVector &	previous_state,
		const SimpleVector &	proposed_state,
		const SimpleVector &	accepted_state
	)

inlinevirtual

template<class SimpleVector>

const kjb::Matrix& Simple_adaptive_mh_step< SimpleVector >::get_cholesky_covariance ( ) const

inline

get cholesky decomposition of covariance matrix before incorporating global scale. This should be multiplied by the global scale to get the actual covariance

template<class SimpleVector>

double Simple_adaptive_mh_step< SimpleVector >::get_current_gamma ( ) const

inline

returns the current learning rate (for debugging)

template<class SimpleVector>

double Simple_adaptive_mh_step< SimpleVector >::get_global_scale ( ) const

inline

get global scaling constant

template<class SimpleVector>

double Simple_adaptive_mh_step< SimpleVector >::get_log_lambda ( ) const

inline

returns the log of the current scaling lambda

template<class SimpleVector>

Step_log<SimpleVector> Simple_adaptive_mh_step< SimpleVector >::operator()	(	SimpleVector &	in,
		double	lt_m
	)

inline

template<class SimpleVector>

void Simple_adaptive_mh_step< SimpleVector >::set_constant_learning_rate	(	double	gamma,
		size_t	change_point = `0`
	)

inline

Use constant learning rate, . When iteration exceeds change_point, adaptation ceases to occur.

Setting change_point to zero causes adaptation to occur indefinitely, but the markov chain will no longer converge to the target distribution. In some situations this is okay, because the convergent distribution is a reasonable approximation to the target distribution, but use with caution.

template<class SimpleVector>

void Simple_adaptive_mh_step< SimpleVector >::set_inverse_learning_rate	(	double	C,
		double	alpha
	)

inline

Let be the adaptation learning rate at iteration i. Calling this function will define the sequence of 's to be:

= C * i^{-} Where >= 0. Recommended values: C<=1, alpha >= 0.5.

Smaller C corresponds to greater confidence in the initial covariance matrix. Increasing alpha corresponds to faster vanishing of the adaptation term, i.e. adaptation will "stop" sooner.

Setting alpha >= 0.5 will ensure that the rate of convergence of the Markov Chain is unaffected by adaptation. See (Andrieu and Thoms, 2008) for details.

template<class SimpleVector>

void Simple_adaptive_mh_step< SimpleVector >::set_post_callback ( const boost::function1< void, const Self & > & cb )

inline

Set a function that will be called at the end of every iteration. Useful for logging/debugging of adaptation

template<class SimpleVector>

void Simple_adaptive_mh_step< SimpleVector >::set_target ( const Target_distribution & t )

inlineprotected

template<class SimpleVector>

void Simple_adaptive_mh_step< SimpleVector >::set_temperature ( double t )

inline

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

/work/kobus/src/lib/sample_cpp/sample_adaptive_mh.h

Public Types

Public Member Functions

Protected Member Functions

Member Typedef Documentation

Constructor & Destructor Documentation

Member Function Documentation