dlmodeler (version 1.4-2)

dlmodeler.filter.smooth: Filtering and smoothing for a DLM

Description

Kalman filtering and smoothing for a Dynamic Linear Model, using the specified back-end for the computations.

Usage

dlmodeler.filter(yt, model, backend = c("KFAS", "FKF", "dlm"), smooth = FALSE, raw.result = FALSE, logLik = FALSE, filter = TRUE)
dlmodeler.smooth(filt, raw.result = FALSE)

Arguments

yt

matrix of observed values (one column per time step).

model

an object of class dlmodeler.

backend

an optional argument which specifies the back-end to use for the computations.

smooth

an optional argument which specifies if the back-end shoud also run the smoothing algorithm.

raw.result

if TRUE, the raw results from the back-end will be stored in raw.result.

logLik

an optional argument which specifies if the back-end shoud compute the log-likelihood.

filter

an optional argument which specifies if the back-end shoud also run the filtering algorithm.

filt

filtered dlmodeler.filtered, as a result from a call to dlmodeler.filter().

Value

An object of class dlmodeler.filtered which contains the following elements:

f

matrix containing the one step ahead predictions $E(y(t) | y(1),y(2)...y(t-1))$

at

matrix containing the one step ahead predicted state variables $E(a(t) | y(1),y(2)...y(t-1))$

Pt

matrix containing the one step ahead predicted state covariance matrices $cov(a(t) | y(1),y(2)...y(t-1))$

logLik

the value of the log-likelihood for the model

backend

a character string indicating which back-end was used for the computation

raw.result

the raw result from the back-end, or NA if it wasn't requested

Or an object of class dlmodeler.smoothed which contains the following elements:

at

matrix containing the one step ahead smoothed state variables $E(a(t) | y(1),y(2)...y(n))$

Pt

matrix containing the one step ahead predicted state covariance matrices $cov(a(t) | y(1),y(2)...y(n))$

backend

a character string indicating which back-end was used for the computation

raw.result

the raw result from the back-end, or NA if it wasn't requested

Details

This function will automatically load the adequate back-end package.

Currently, packages KFAS (used by default), FKF and dlm are supported. Refer to dlmodeler for more information.

See Also

dlmodeler, dlmodeler.forecast

Examples

## Not run: 
# require(dlmodeler)
# 
# # generate some data
# N <- 365*5
# t <- c(1:N,rep(NA,365))
# a <- rnorm(N+365,0,.5)
# y <- pi + cos(2*pi*t/365.25) + .25*sin(2*pi*t/365.25*3) +
#      exp(1)*a + rnorm(N+365,0,.5)
# 
# # build a model for this data
# m <- dlmodeler.build.polynomial(0,sigmaH=.5,name='level') +
#      dlmodeler.build.dseasonal(7,sigmaH=0,name='week') +
#      dlmodeler.build.tseasonal(365.25,3,sigmaH=0,name='year') +
#      dlmodeler.build.regression(a,sigmaH=0,name='reg')
# m$name <- 'mymodel'
# 
# system.time(f <- dlmodeler.filter(y, m, raw.result=TRUE))
# 
# # extract all the components
# m.state.mean <- dlmodeler.extract(f,m,type="state",
#                                   value="mean")
# m.state.cov <- dlmodeler.extract(f,m,type="state",
#                                  value="covariance")
# m.obs.mean <- dlmodeler.extract(f,m,type="observation",
#                                 value="mean")
# m.obs.cov <- dlmodeler.extract(f,m,type="observation",
#                                value="covariance")
# m.obs.int <- dlmodeler.extract(f,m,type="observation",
#                                value="interval",prob=.99)
# 
# par(mfrow=c(2,1))
# 
# # show the one step ahead forecasts & 99% prediction intervals
# plot(y,xlim=c(N-10,N+30))
# lines(m.obs.int$mymodel$upper[1,],col='light grey')
# lines(m.obs.int$mymodel$lower[1,],col='light grey')
# lines(m.obs.int$mymodel$mean[1,],col=2)
# 
# # see to which values the filter has converged:
# m.state.mean$level[,N] # should be close to pi
# mean(abs(m.state.mean$week[,N])) # should be close to 0
# m.state.mean$year[1,N] # should be close to 1
# m.state.mean$year[6,N] # should be close to .25
# m.state.mean$reg[,N] # should be close to e
# 
# # show the filtered level+year components
# plot(m.obs.mean$level[1,]+m.obs.mean$year[1,],
# 		type='l',ylim=c(pi-2,pi+2),col='light green',
# 		ylab="smoothed & filtered level+year")
# 
# system.time(s <- dlmodeler.smooth(f,m))
# 
# # show the smoothed level+year components
# s.obs.mean <- dlmodeler.extract(s,m,type="observation",
#                                 value="mean")
# lines(s.obs.mean$level[1,]+s.obs.mean$year[1,],type='l',
# 		ylim=c(pi-2,pi+2),col='dark green')
# ## End(Not run)