Filter< A, TransKernel, I, O, Ts > Struct Template Reference

A generic filtering inference algorithm. More...

#include "filter.hpp"

Inheritance diagram for Filter< A, TransKernel, I, O, Ts >:

Public Member Functions
	Filter (pp_t< I, O, Ts... > f, FilterQueryerType< O, Ts... > &queryer, transition< TransKernel, O, Ts... > &trans, inf_options_t opts, size_t seed=2022)
	Construct a new Filter object. More...
template<template< typename, typename, typename, template< class, class, class... > class, typename... > class G = A>
std::enable_if_t< has_proposal< G >::value==kernel::none, FilterValueType< O, Ts... > >	step (I &input)
	Executes one filtering step. More...
template<class P , template< typename, typename, typename, template< class, class, class... > class, typename... > class G = A>
std::enable_if_t< has_proposal< G >::value==kernel::full, FilterValueType< O, Ts... > >	step (I &input, P &proposal)
	Executes one filtering step. More...
void	update_internal_state (FilterValueType< O, Ts... > &posterior)

Public Attributes
bool	has_been_called
std::minstd_rand	rng
transition< TransKernel, O, Ts... > &	trans

Additional Inherited Members
Protected Member Functions inherited from Inference< A, I, O, FilterValueType< O, Ts... >, WeightedRecord, Ts... >
	Inference (pp_t< I, O, Ts... > f, WeightedRecord< FilterValueType< O, Ts... >, O, Ts... > &queryer, inf_options_t opts)
std::enable_if_t< has_proposal< G >::value==kernel::none, FilterValueType< O, Ts... > >	operator() (I &input)
	Runs the specified inference algorithm with the specified queryer. More...
std::enable_if_t< has_proposal< G >::value==kernel::full, FilterValueType< O, Ts... > >	operator() (I &input, P &proposal)
	Runs the specified inference algorithm with the specified queryer using the specified proposal distribution. More...
std::enable_if_t< has_proposal< G >::value==kernel::partial, FilterValueType< O, Ts... > >	operator() (I &input, pp_t< record_t< Ts... >, double, Ts... > &proposal)
	Runs the specified inference algorithm with the specified queryer using the specified proposal distribution. More...
Protected Attributes inherited from Inference< A, I, O, FilterValueType< O, Ts... >, WeightedRecord, Ts... >
pp_t< I, O, Ts... >	f
inf_options_t	opts
WeightedRecord< FilterValueType< O, Ts... >, O, Ts... > &	queryer

Detailed Description

template<template< typename, typename, typename, template< class, class, class... > class, typename... > class A, class TransKernel, typename I, typename O, typename... Ts>
struct Filter< A, TransKernel, I, O, Ts >

A generic filtering inference algorithm.

Given a probabilistic program \(p(x,z) = p(x|z)p(z)\) and a user-supplied transition kernel, the model is converted into a dynamic filtering model \( p(x,z) = p(x_1 | z_1) p(z_1) \prod_{t\geq 2} p(x_t | z_t)p(z_t|x_{t-1}, z_{t-1})\) using filter_effect. At each timestep, inference is performed using the user-specified inference algorithm to recover an approximate posterior \(p(z_t | x_t)\). That approximate posterior is consumed by the transition callable to generate the new transition dynamics \(p(z_{t+1}|x_t, z_t)\) used in the next filtering step.

Template Parameters

A	The class template of the inference algorithm to use in constructing the approximate posterior
I	The input type of the probabilistic program
O	The output type of the probabilistic program
Ts	The types of the distributions in the probabilistic program

---

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

• lib/inference/filter.hpp