siskernel: Sequential Importance Sampling with arbitrary kernel for 1-D...

In nickpoison/nltsa: Nonlinear Time Series Analysis

Description

Runs sequential importance sampling without resampling on a given Non-Linear State Space models with user-specified kernel as proposal.

Usage

  siskernel(nlss, y, N, proposal.rnd = prior.rnd,
    proposal.logpdf = prior.logpdf,
    initial.proposal.rnd = initial.rnd,
    initial.proposal.logpdf = initial.logpdf)

Arguments

nlss	Non-linear state space model
y	Sequence of observations. Its length T is the number of timesteps.
N	Number of particles
proposal.rnd	Function sampling from the proposal kernel to use
proposal.logpdf	Function computing the log-pdf of the proposal kernel
initial.proposal.rnd	Function sampling from the proposal kernel to use at initial timestep
initial.proposal.logpdf	Function computing the log-pdf of the proposal kernel at initial timestep

Details

This algorithm is a slightly more generic version of sis. It is not recommended and included for illustrative purposes only. This version is therefore a minimalistic and only supports NLLS with univariate states. Use sisr instead.

Value

A list with the following components:

particles	Array (T, N, D) of the sampled particles
logweights	Array (T, N) of the logarithm of the non-normalized importance weights of the particles
weights	Array (T, N) of the normalized importance weights of the particles
t	Indices 1 to T, included for ease of plotting

See Also

sisr

nickpoison/nltsa documentation built on May 23, 2019, 4:48 p.m.