GibbsDyn: MCMC sampling for the models.

In spTDyn: Spatially Varying and Spatio-Temporal Dynamic Linear Models

MCMC sampling for the models.

Description

This function is used to draw MCMC samples using the Gibbs sampler.

Usage

GibbsDyn(formula, data=parent.frame(), model="GP", time.data=NULL, coords, 
	priors=NULL, initials=NULL, nItr=5000, nBurn=1000, report=1, tol.dist=0.05, 
	distance.method="geodetic:km", cov.fnc="exponential", scale.transform="NONE", 
	spatial.decay=decay(distribution="FIXED"),truncation.para=list(at=0,lambda=2))

Arguments

formula	The symnbolic description of the model equation of the regression part of the space-time model. The terms sp and tp are used to define spatially and temporally varying parameters for the model.
data	An optional data frame containing the variables in the model. If omitted, the variables are taken from environment(formula), typically the environment from which spT.Gibbs is called. The data should be ordered first by the time and then by the sites specified by the coords below. One can also supply coordinates through this argument, where coordinate names should be "Latitude" and "Longitude".
model	The spatio-temporal models to be fitted, current choices are: "GP", and "truncated", with the first one as the default.
time.data	Defining the segments of the time-series set up using the function def.time.
coords	The n by 2 matrix or data frame defining the locations (e.g., longitude/easting, latitude/northing) of the fitting sites, where n is the number of fitting sites. One can also supply coordinates through a formula argument such as ~Longitude+Latitude.
priors	The prior distributions for the parameters. Default distributions are specified if these are not provided. If priors=NULL a flat prior distribution will be used with large variance. See details in priors.
initials	The preferred initial values for the parameters. If omitted, default values are provided automatically. Further details are provided in initials.
nItr	Number of MCMC iterations. Default value is 5000.
nBurn	Number of burn-in samples. This number of samples will be discarded before making any inference. Default value is 1000.
report	Number of reports to display while running the Gibbs sampler. Defaults to number of iterations.
distance.method	The preferred method to calculate the distance between any two locations. The available options are "geodetic:km", "geodetic:mile", "euclidean", "maximum", "manhattan", and "canberra". See details in dist. The default is "geodetic:km".
tol.dist	Minimum separation distance between any two locations out of those specified by coords, knots.coords and pred.coords. The default is 0.005. The programme will exit if the minimum distance is less than the non-zero specified value. This will ensure non-singularity of the covariance matrices.
cov.fnc	Covariance function for the spatial effects. The available options are "exponential", "gaussian", "spherical" and "matern". If "matern" is used then by default the smooth parameter (ν) is estimated from (0,1) uniform distribution using discrete samples.
scale.transform	The transformation method for the response variable. Currently implemented options are: "NONE", "SQRT", and "LOG" with "NONE" as the deault.
spatial.decay	Provides the prior distribution for the spatial decay parameter φ. Currently implemented options are "FIXED", "Unif", or "Gamm". Further details for each of these are specified by decay.
truncation.para	Provides truncation parameter λ and truncation point "at" using list.

Value

accept	The acceptance rate for the φ parameter if the "MH" method of sampling is chosen.
phip	MCMC samples for the parameter φ.
nup	MCMC samples for the parameter ν. Only available if "matern" covariance function is used.
sig2eps	MCMC samples for the parameter σ^2_ε.
sig2etap	MCMC samples for the parameter σ^2_η.
sig2betap	MCMC samples for the parameter σ^2_β, only applicable for spatially varying coefficient process model.
sig2deltap	MCMC samples for the parameter σ^2_δ, for β_j, j=1,...,u. Only applicable for spatio-temporal DLM.
sig2op	MCMC samples for the parameter σ^2_o, for initial variance of β_0. Only applicable for spatio-dynamic and spatio-temporal DLM.
betap	MCMC samples for the parameter β.
rhop	MCMC samples for ρ.
op	MCMC samples for the true observations.
fitted	MCMC summary (mean and sd) for the fitted values.
tol.dist	Minimum tolerance distance limit between the locations.
distance.method	Name of the distance calculation method.
cov.fnc	Name of the covariance function used in model fitting.
scale.transform	Name of the scale.transformation method.
sampling.sp.decay	The method of sampling for the spatial decay parameter φ.
covariate.names	Name of the covariates used in the model.
Distance.matrix	The distance matrix.
coords	The coordinate values.
n	Total number of sites.
r	Total number of segments in time, e.g., years.
T	Total points of time, e.g., days within each year.
p	Total number of model coefficients, i.e., β's including the intercept.
initials	The initial values used in the model.
priors	The prior distributions used in the model.
PMCC	The predictive model choice criteria obtained by minimising the expected value of a loss function, see Gelfand and Ghosh (1998). Results for both goodness of fit and penalty are given.
iterations	The number of samples for the MCMC chain, without burn-in.
nBurn	The number of burn-in period for the MCMC chain.
computation.time	The computation time required for the fitted model.

References

Bakar, K. S., Kokic, P. and Jin, H. (2015). A spatio-dynamic model for assessing frost risk in south-eastern Australia. Journal of the Royal Statistical Society, Series C. Bakar, K. S., Kokic, P. and Jin, H. (2015). Hierarchical spatially varying coefficient and temporal dynamic process models using spTDyn. Journal of Statistical Computation and Simulation.

See Also

priors, initials, dist, sp, tp.

Examples

##

###########################
## Attach library spTDyn
###########################

library(spTDyn)

## Read Aus data ##
data(AUSdata)
# set a side data for validation
library(spTimer)
s<-c(1,4,10)
AUSdataFit<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s, reverse=TRUE)
AUSdataFit<-subset(AUSdataFit, with(AUSdataFit, !(year == 2009)))
AUSdataPred<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s)
AUSdataPred<-subset(AUSdataPred, with(AUSdataPred, !(year == 2009)))
AUSdataFore<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s)
AUSdataFore<-subset(AUSdataFore, with(AUSdataFore, (year == 2009)))

## Read NY data ##
data(NYdata)
# set a side data for validation
s<-c(5,8,10,15,20,22,24,26)
fday<-c(25:31)
NYdataFit<-spT.subset(data=NYdata, var.name=c("s.index"), s=s, reverse=TRUE)
NYdataFit<-subset(NYdataFit, with(NYdataFit, !(Day %in% fday & Month == 8)))
NYdataPred<-spT.subset(data=NYdata, var.name=c("s.index"), s=s)
NYdataPred<-subset(NYdataPred, with(NYdataPred, !(Day %in% fday & Month == 8)))
NYdataFore<-spT.subset(data=NYdata, var.name=c("s.index"), s=s)
NYdataFore<-subset(NYdataFore, with(NYdataFore, (Day %in% fday & Month == 8)))

## Code for analysing temperature data in Section: 4 ##
## Model: Spatially varying coefficient process models ##

nItr<-13000
nBurn<-3000

# MCMC via Gibbs using defaults
# Spatially varying coefficient process model

library("spTDyn", warn.conflicts = FALSE)
set.seed(11)
post.sp <- GibbsDyn(tmax ~ soi+sp(soi)+grid+sp(grid),
           data=AUSdataFit, nItr=nItr, nBurn=nBurn, coords=~lon+lat,
           spatial.decay=decay(distribution=Gamm(2,1),tuning=0.06))
print(post.sp)

## Table: 3, Section: 4.1 ##
post.sp$PMCC

# parameter summary
summary(post.sp) # without spatially varying coefficients
summary(post.sp, coefficient="spatial")

#plot(post.sp, density=FALSE)  # without spatially varying coefficients
#plot(post.sp, coefficient="spatial", density=FALSE)

## Code for Figures: 3(a), 3(b) Section: 4.1 ##
Figure_3a<-function(){
  boxplot(t(post.sp$betasp[1:9,]),pch=".",main="SOI",
          xlab="Sites",ylab="Values")
}
Figure_3b<-function(){
  boxplot(t(post.sp$betasp[10:18,]),pch=".",main="Grid",
          xlab="Sites",ylab="Values")
}
Figure_3a()
Figure_3b()

## spatial prediction
set.seed(11)
pred.sp <- predict(post.sp,newcoords=~lon+lat,newdata=AUSdataPred)

## Table: 4, Section: 4.1, validations ##
spT.validation(AUSdataPred$tmax,c(pred.sp$Mean))
plot(AUSdataPred$tmax,c(pred.sp$Mean))

## temporal prediction
set.seed(11)
pred.sp.f <- predict(post.sp,type="temporal",foreStep=12,
                     newcoords=~lon+lat, newdata=AUSdataFore)

## Table: 4, Section: 4.1, validations ##
spT.validation(AUSdataFore$tmax,c(pred.sp.f$Mean))
plot(AUSdataFore$tmax,c(pred.sp.f$Mean))

## Code for analysing Ozone data in Section: 4 ##
## Model: spatio-temporal DLM ##

# MCMC via Gibbs using defaults
# spatio-temporal DLM

library("spTDyn", warn.conflicts = FALSE)
set.seed(11)
post.tp <- GibbsDyn(o8hrmax ~ tp(cMAXTMP)-1, data=NYdataFit,
           nItr=nItr, nBurn=nBurn, coords=~Longitude+Latitude,
           initials=initials(rhotp=0), scale.transform="SQRT",
           spatial.decay=decay(distribution=Gamm(2,1),tuning=0.05))
print(post.tp)
summary(post.tp)

## Table: 5, Section: 4.2 ##
post.tp$PMCC

## Figure: 5, Section: 4.2 ##
Figure_5<-function(){
  stat<-apply(post.tp$betatp[1:55,],1,quantile,prob=c(0.025,0.5,0.975))
  plot(stat[2,],type="p",lty=3,col=1,ylim=c(min(c(stat)),max(c(stat))),
       pch=19,ylab="",xlab="Days",axes=FALSE,main="cMAXTMP",cex=0.8)
  for(i in 1:55){
    segments(i, stat[2,i], i, stat[3,i])
    segments(i, stat[2,i], i, stat[1,i])
  }
  axis(1,1:55,labels=1:55);axis(2)
  abline(v=31.5,lty=2)
  text(15,0.32,"July");  text(45,0.32,"August");
}
Figure_5()

## spatial prediction
set.seed(11)
pred.tp <- predict(post.tp, newdata=NYdataPred, newcoords=~Longitude+Latitude)

## Table 6, Section: 4.2, validation ##
spT.validation(NYdataPred$o8hrmax,c(pred.tp$Mean))

## temporal prediction
set.seed(11)
pred.tp.f <- predict(post.tp, newdata=NYdataFore, newcoords=~Longitude+Latitude,
                     type="temporal", foreStep=7)

## Table 6, Section: 4.2, validation ##
spT.validation(NYdataFore$o8hrmax,c(pred.tp.f$Mean))

######################################################
## The Truncated/Censored models:
######################################################

## Read Aus data ##
data(AUSdata)
# set the truncation point at tmax=30
AUSdata$tmax <- replace(AUSdata$tmax, AUSdata$tmax<=30, 30)

# set a side data for validation
library(spTimer)
s<-c(1,4,10)
AUSdataFit<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s, reverse=TRUE)
AUSdataFit<-subset(AUSdataFit, with(AUSdataFit, !(year == 2009)))
AUSdataPred<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s)
AUSdataPred<-subset(AUSdataPred, with(AUSdataPred, !(year == 2009)))
AUSdataFore<-spT.subset(data=AUSdata, var.name=c("s.index"), s=s)
AUSdataFore<-subset(AUSdataFore, with(AUSdataFore, (year == 2009)))

#
nItr <- 5000 # number of MCMC samples for each model
nBurn <- 1000 # number of burn-in from the MCMC samples
# Truncation at 30 
# fit truncated spatially varying model 

## The Truncated/Censored spatially varying models:
library("spTDyn", warn.conflicts = FALSE)
set.seed(11)
out <- GibbsDyn(tmax ~ soi+sp(soi)+grid+sp(grid),model="truncated",
           data=AUSdataFit, nItr=nItr, nBurn=nBurn, coords=~lon+lat,
           spatial.decay=decay(distribution=Gamm(2,1),tuning=0.06),
           truncation.para = list(at = 30,lambda = 2))
print(out)
summary(out)
head(fitted(out))
plot(out,density=FALSE)
#
head(cbind(AUSdataFit$tmax,fitted(out)[,1]))
plot(AUSdataFit$tmax,fitted(out)[,1])
spT.validation(AUSdataFit$tmax,fitted(out)[,1])

## spatial prediction
set.seed(11)
pred.sp <- predict(out,newcoords=~lon+lat,newdata=AUSdataPred)
spT.validation(AUSdataPred$tmax,c(pred.sp$Mean))
plot(AUSdataPred$tmax,c(pred.sp$Mean))

## temporal prediction
set.seed(11)
pred.sp.f <- predict(out,type="temporal",foreStep=12,
                     newcoords=~lon+lat, newdata=AUSdataFore)
spT.validation(AUSdataFore$tmax,c(pred.sp.f$Mean))
plot(AUSdataFore$tmax,c(pred.sp.f$Mean))

## The Truncated/Censored temporal dynamic DLM models:
library("spTDyn", warn.conflicts = FALSE)
set.seed(11)
out <- GibbsDyn(tmax ~ soi+tp(soi)+grid,model="truncated",
           data=AUSdataFit, nItr=nItr, nBurn=nBurn, coords=~lon+lat,
           spatial.decay=decay(distribution=Gamm(2,1),tuning=0.06),
           truncation.para = list(at = 30,lambda = 2))
print(out)
summary(out)
head(fitted(out))
plot(out,density=FALSE)
#
head(cbind(AUSdataFit$tmax,fitted(out)[,1]))
plot(AUSdataFit$tmax,fitted(out)[,1])
spT.validation(AUSdataFit$tmax,fitted(out)[,1])

## spatial prediction
set.seed(11)
pred.tp <- predict(out,newcoords=~lon+lat,newdata=AUSdataPred)
spT.validation(AUSdataPred$tmax,c(pred.tp$Mean))
plot(AUSdataPred$tmax,c(pred.tp$Mean))

## temporal prediction
set.seed(11)
pred.tp.f <- predict(out,type="temporal",foreStep=12,
                     newcoords=~lon+lat, newdata=AUSdataFore)
spT.validation(AUSdataFore$tmax,c(pred.tp.f$Mean))
plot(AUSdataFore$tmax,c(pred.tp.f$Mean))


##############################################################################

spTDyn documentation built on Nov. 22, 2022, 5:07 p.m.