f.RdFunction used for defining of smooth and spatial terms within inla model
formulae. The function does not evaluate anything - it
exists purely to help set up a model. The function specifies one
smooth function in the linear predictor (see inla.list.models) as
$$w\ f(x)$$
f(...,
model = "iid",
copy=NULL,
same.as = NULL,
n=NULL,
nrep = NULL,
replicate = NULL,
ngroup = NULL,
group = NULL,
control.group = inla.set.control.group.default(),
hyper = NULL,
initial=NULL,
prior=NULL,
param = NULL,
fixed = NULL,
season.length=NULL,
constr = NULL,
extraconstr=list(A=NULL, e=NULL),
values=NULL,
cyclic = NULL,
diagonal = NULL,
graph=NULL,
graph.file=NULL,
cdf=NULL,
quantiles=NULL,
Cmatrix=NULL,
rankdef=NULL,
Z = NULL,
nrow = NULL,
ncol = NULL,
nu = NULL,
bvalue = NULL,
spde.prefix = NULL,
spde2.prefix = NULL,
spde2.transform = c("logit", "log", "identity"),
spde3.prefix = NULL,
spde3.transform = c("logit", "log", "identity"),
mean.linear = inla.set.control.fixed.default()$mean,
prec.linear = inla.set.control.fixed.default()$prec,
compute = TRUE,
of=NULL,
precision = exp(14),
range = NULL,
adjust.for.con.comp = TRUE,
order = NULL,
scale = NULL,
strata = NULL,
rgeneric = NULL,
scale.model = NULL,
args.slm = list(rho.min = NULL, rho.max = NULL,
X = NULL, W = NULL, Q.beta = NULL),
args.ar1c = list(Z = NULL, Q.beta = NULL),
args.intslope = list(subject = NULL, strata = NULL, covariates = NULL),
correct = NULL,
vb.correct = NULL,
locations = NULL,
debug = FALSE)Name of the covariate and, possibly of the weights vector. NB: order counts!!!! The first specified term is the covariate and the second one is the vector of weights (which can be negative).
A string indicating the choosen model. The
default is iid. See
names(inla.models()$latent) for a list of possible
alternatives and inla.doc for detailed docs.
TODO
TODO
An optional argument which defines the dimension
of the model if this is different from
length(sort(unique(covariate)))
TODO
We need to write documentation here
TODO
TODO
TODO
Specification of the hyperparameter, fixed or
random, initial values, priors and its parameters. See
?inla.models for the list of hyparameters for each
model and its default options or
use inla.doc() for
detailed info on the family and
supported prior distributions.
THIS OPTION IS OBSOLETE; use
hyper!!! Vector indicating the starting values for
the optimization algorithm. The length of the vector
depends on the number of hyperparamters in the choosen
model. If fixed=T the value at which the
parameters are fixed is determines through initial.
See inla.models()$latent$'model name' to have info
about the choosen model.
THIS OPTION IS OBSOLETE; use hyper!!!
Prior distribution(s) for the hyperparameters of the
!random model. The default value depends on the type of
model, see !www.r-inla.org for a detailed
description of the models. See
names(inla.models()$priors) for possible prior
choices
THIS OPTION IS OBSOLETE; use hyper!!!
Vector indicating the parameters \(a\) and \(b\)
of the prior distribution for the hyperparameters. The
length of the vector depends on the choosen model.
See inla.models()$latent$'model name' to have info
about the choosen model.
THIS OPTION IS OBSOLETE; use hyper!!!
Vector of boolean variables indicating wheater the
hyperparameters of the model are fixed or random. The
length of the vector depends on the choosen model
See inla.models()$latent$'model name' to have info
about the choosen model.
Lenght of the seasonal compoment
(ONLY if model="seasonal")
A boolean variable indicating whater to set a sum to 0 constraint on the term. By default the sum to 0 constraint is imposed on all intrinsic models ("iid","rw1","rw1","besag", etc..).
This argument defines extra linear
constraints. The argument is a list with two elements, a
matrix A and a vector e, which defines the
extra constraint Ax = e; for example
extraconstr = list(A = A, e=e). The number of
columns of A must correspond to the length of this
f-model. Note that this constraint comes
additional to the sum-to-zero constraint defined if
constr = TRUE.
An optional vector giving all values
assumed by the covariate for which we want estimated the
effect. It must be a numeric vector, a vector of factors
or NULL.
A boolean specifying wheather the model is cyclical. Only valid for "rw1" and "rw2" models, is cyclic=T then the sum to 0 constraint is removed. For the correct form of the grah file see Martino and Rue (2008).
An extra constant added to the diagonal of the precision matrix.
Defines the graph-object either as a file with
a graph-description, an inla.graph-object, or as a
(sparse) symmetric matrix.
THIS OPTION IS OBSOLETE AND REPLACED BY
THE MORE GENERAL ARGUMENT graph. PLEASE CHANGE YOUR
CODE.
Name of the file containing the graph
of the model; see
www.r-inla.org/faq.
A vector of maximum 10 values between 0 and 1 \(x(0), x(1),\ldots\). The function returns, for each posterior marginal the probabilities $$\mbox{Prob}(X<x(p))$$
A vector of maximum 10 quantiles, \(p(0), p(1),\dots\) to compute for each posterior marginal. The function returns, for each posterior marginal, the values \(x(0), x(1),\dots\) such that $$\mbox{Prob}(X<x(p))=p$$
The specification of the precision matrix
for the generic, generic3 or z models (up to a scaling constant).
Cmatrix is either a
(dense) matrix, a matrix created using
Matrix::sparseMatrix(), or a filename which stores the
non-zero elements of Cmatrix, in three columns:
i, j and Qij. In case of the generic3 model,
it is a list of such specifications.
A number defining the rank deficiency of the model, with sum-to-zero constraint and possible extra-constraints taken into account. See details.
The matrix for the z-model
Number of rows for 2d-models
Number of columns for 2d-models
Smoothing parameter for the Matern2d-model,
possible values are c(0, 1, 2, 3)
TODO
TODO
TODO
TODO
TODO
TODO
Prior mean for the linear component,
only used if model="linear"
Prior precision for the linear
component, only used if model="linear"
A boolean variable indicating wheather the
marginal posterior distribution for the nodes in the
f() model should be computed or not. This is
usefull for large models where we are only interested in
some posterior marginals.
TODO
The precision for the artifical noise added when creating a copy of a model and others.
A vector of size two giving the lower and
upper range for the scaling parameter beta in the
model COPY, CLINEAR, MEC and MEB.
If low = high then the identity mapping
is used.
If TRUE (default), adjust some of the models (currently: besag, bym, bym2 and besag2) if the number of connected components in graph is larger than 1. If FALSE, do nothing.
Defines the order of the model: for
model ar this defines the order p, in AR(p). Not
used for other models at the time being.
A scaling vector. Its meaning depends on the model.
Currently not in use
A object of class inla.rgeneric which defines the model. (EXPERIMENTAL!)
Logical. If TRUE then scale the RW1 and RW2 and BESAG and BYM and BESAG2 and RW2D models so the their (generlized) variance is 1. Default value is inla.getOption("scale.model.default")
Required arguments to the model="slm"; see the documentation for further details.
,
,
A list with the subject (factor), strata (factor) and covariates (numeric) for the intslope model; see the documentation for further details.
,
Add this model component to the list of variables to be used in the corrected Laplace approximation? If NULL use default choice, otherwise correct if TRUE and do not if FALSE. (This option is currently experimental.)
,
Add this model component to the list of variables to be used for the vb corrected Laplace approximation? If NULL use default choice, otherwise correct if TRUE and do not if FALSE. (expert option)
,
TODO
There is no default value for rankdef, if it
is not defined by the user then it is computed by the rank
deficiency of the prior model (for the generic model, the
default is zero), plus 1 for the sum-to-zero constraint if the
prior model is proper, plus the number of extra
constraints. Oops: This can be wrong, and then the user
must define the rankdef explicitely.
inla, hyperpar.inla