Description Usage Arguments Details Value Note Author(s) References See Also Examples
Builds an univariate (multi-linear) regression DLM.
1 2 3 4 5 | dlmodeler.regression(covariates, sigmaH = NA, sigmaQ = 0,
intercept = FALSE, name = "regression")
dlmodeler.build.regression(covariates, sigmaH = NA, sigmaQ = 0,
intercept = FALSE, name = "regression")
|
covariates |
covariate matrix (one row per covariate). |
sigmaH |
std dev of the observation disturbance (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = NA. |
sigmaQ |
std dev of the state disturbance (if unknown, set to NA and use dlmodeler.fit to estimate it). Default = 0. |
intercept |
should an intercept be added to the model? |
name |
an optional name to be given to the resulting DLM. |
The regression term is of the form a[1] x_1(t) + a[2]x_2(t) ... + a[k]x_k(t), where x_k is the k-th covariate.
The initial value P0inf
is parametered to use exact diffuse initialisation
(if supported by the back-end).
An object of class dlmodeler
representing the regression model.
State representations are not unique, so other forms could be used to achieve the same goals.
Cyrille Szymanski <cnszym@gmail.com>
Durbin, and Koopman, Time Series Analysis by State Space Methods, Oxford University Press (2001), pages 38-45.
dlmodeler
,
dlmodeler.build
,
dlmodeler.build.polynomial
,
dlmodeler.build.dseasonal
,
dlmodeler.build.tseasonal
,
dlmodeler.build.structural
,
dlmodeler.build.arima
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | ## 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)
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.