PFSMC: Sequential Monte Carlo Partical Filter

In azure10h/PFSMC:

Description

A kinetic prediction implemented with Sequential Monte Carlo algorithm

Usage

PFSMC(Y, eta, alpha, N, c, T, loss, resample = resampleMultinomial)

Arguments

Y	a 1*T vector of sequential data sequence.
eta	learning parameter. Determines the rate of weight updating process.
alpha	mixing parameter. Determines the speed of model convergence and the rate of trakcing changepoints.
N	number of particles to predict underlying densities.
c	effective sample size thershold.
T	number of data sequence. Time index.
loss	loss function for the underlying density. Used to update sample weights.
resample	resampling function.

Value

PFSMC returns a list of effective sample size, normalized constants, predicted parameters theta and resample flags at each time.

Examples

#Generate true parameters and data sequence with 5 change points
k=5; T=200
library(PFSMC)
Data=datagenNoml(200,5,-10,10)
Y=Data[[1]]
theta_true=Data[[2]]

#Detecting changepoints using `PFSMC` funciton.
#We choose a score function for Gaussian distribution and 
a multinomial resampling method.
Simulation<-PFSMC(Y=Y,eta=10*T^(-1/3),alpha=k/(T-1),N=1000,
c=0.5,T=200,loss= lossGaussian, resample=resampleMultinomial)
ESS=Simulation[[1]]
theta_hat=Simulation$theta_hat

#Result visulization
plot(theta_true, type="l", ylim=c(-12,12), xlab="Time", ylab="Predictive/True Parameters")
lines(theta_hat, col="red")

azure10h/PFSMC documentation built on Feb. 5, 2020, 5:11 a.m.