lppl
v2.0.0
|
Computes a variational approximation to posterior from (mean, standard deviation) pairs. More...
#include "update_impls.hpp"
Public Types | |
using | typed_map_ = typed_map< Policy, I, Ts... > |
Public Types inherited from Update< ParameterMatching, Policy, ProductGenerator< WeightedMeanStd, std::pair< double, double >, O, Ts... >::EmitType, I, O, Ts... > | |
using | typed_map_ = typed_map< Policy, I, Ts... > |
Public Member Functions | |
typed_map_ | operator() (ProductGenerator< WeightedMeanStd, std::pair< double, double >, O, Ts... >::EmitType &result) |
Public Member Functions inherited from Update< ParameterMatching, Policy, ProductGenerator< WeightedMeanStd, std::pair< double, double >, O, Ts... >::EmitType, I, O, Ts... > | |
typed_map_ | operator() (ProductGenerator< WeightedMeanStd, std::pair< double, double >, O, Ts... >::EmitType &result) |
Calls derived implementation. More... | |
Additional Inherited Members | |
Public Attributes inherited from Update< ParameterMatching, Policy, ProductGenerator< WeightedMeanStd, std::pair< double, double >, O, Ts... >::EmitType, I, O, Ts... > | |
upp_t< Policy, I, O, Ts... > & | f |
Computes a variational approximation to posterior from (mean, standard deviation) pairs.
Given samples from the posterior (e.g., as computed by importance sampling or MCMC methods), parameter_matching
computes a factorized functional approximation to the posterior suitable for use in message passing or as a new prior distribution for filtering applications.
parameter_matching
assumes that the posterior factors as \(p(z_1,...|x) = \prod_{n \geq 1} p_{\psi_n'}(z_n)\), where each component of the density has the same functional form as its counterpart in the prior. The parameters are defined by \(\psi_n' = f(\mu, \sigma)\), where the definition of the mapping \(f\) from the mean \(\mu\) and standard deviation \(\sigma\) are left up to the implementation.
Policy | |
I | |
O | |
Ts |