| Distribution Families {RandomFields} | R Documentation |
Distribution families to specify random parameters in the model definition.
See Bayesian Modelling for some less technical introduction to hierarchical modelling.
When simulating Gaussian random fields, the random parameters
are drawn only once at the very beginning.
So, if the argument n in RFsimulate
is greater than 1 then n simulations conditional
on a single realisation of the random parameters are
performed. See the examples below.
There are (simple) multivariate version and additional version to the distributions families implemented. Further, any distribution family defined in R can be used, see the examples below
These function will allow for Bayesian modelling. (Future project).
RRdeterm | no scattering |
RRdistr | families of distributions transferred from R |
RRgauss | a (multivariate) Gaussian random variable |
RRloc | modification of location and scale |
RRspheric | random scale for
the RMball to simulate RMspheric, etc.
|
RRunif | a (multivariate) uniform random variable |
The allowance of random parameters is a very recent, developing feature of RandomField.
Future changings of the behaviour are not unlikely.
A further random element is RMsign, which is an
operator on shape functions. As an exception its name starts with
RM and not with RR.
Martin Schlather, schlather@math.uni-mannheim.de http://ms.math.uni-mannheim.de/de/publications/software
RC, RF, RM, RP, R.
Other models,
RFdistr,
RMmodelgenerator,
RFoptions(seed=0) ## *ANY* simulation will have the random seed 0; set
## RFoptions(seed=NA) to make them all random again
## here, the scale is given by an exponential variable:
model <- RMgauss(scale=exp())
for (i in 1:4) {
RFoptions(seed = i)
# each leads to a simulation with a different scale parameter
plot(model) ## random
plot(RFsimulate(model, x=seq(0,10,0.1)))
readline("press return")
}
# but here, all 4 simulations have same (but random) scale:
plot(RFsimulate(model, x=seq(0,10,0.1), n=4))
## hierarchical models are also possible:
## here, the scale is given by an exponential variable whose
## rate is given by a uniform variable
model <- RMgauss(scale=exp(rate=unif()))
plot(model)
plot(RFsimulate(model, x=seq(0,10,0.1)))
## HOWEVER, the next model is deterministic with scale \eqn{e=2.718282}.
model <- RMgauss(scale=exp(1))
plot(model)
plot(RFsimulate(model, x=seq(0,10,0.1)))