R/set.default.arguments.R
In andrewzm/INLA: Functions which allow to perform full Bayesian analysis of latent Gaussian models using Integrated Nested Laplace Approximaxion
## Export: inla.set.control.lincomb.default
## Export: inla.set.control.update.default
## Export: inla.set.control.group.default
## Export: inla.set.control.mix.default
## Export: inla.set.control.link.default
## Export: inla.set.control.expert.default
## Export: inla.set.control.compute.default
## Export: inla.set.control.family.default
## Export: inla.set.control.fixed.default
## Export: inla.set.control.inla.default
## Export: inla.set.control.predictor.default
## Export: inla.set.control.results.default
## Export: inla.set.control.mode.default
## Export: inla.set.control.hazard.default
## Export: control.lincomb
## Export: control.update
## Export: control.group
## Export: control.mix
## Export: control.link
## Export: control.expert
## Export: control.compute
## Export: control.family
## Export: control.fixed
## Export: control.inla
## Export: control.predictor
## Export: control.results
## Export: control.mode
## Export: control.hazard
### Defines default arguments
`inla.set.control.update.default` =
function(...)
{
##:EXTRA:
##:NAME: control.update
list(
##:ARGUMENT: result Update the joint posterior for the hyperparameters from result
result = NULL
)
##:SEEALSO: inla
}
`inla.set.control.lincomb.default` =
function(...)
{
##:EXTRA:
##:NAME: control.lincomb
list(
##:ARGUMENT: precision The precision for the artificial tiny noise. Default 1e09.
precision = 10^9,
##:ARGUMENT: verbose Use verbose mode for linear combinations if verbose model is set globally. (Default TRUE)
verbose = TRUE)
##:SEEALSO: inla
}
`inla.set.control.group.default` =
function(...)
{
##:EXTRA:
##:NAME: control.group
list(
##:ARGUMENT: model Group model (one of 'exchangable', 'ar1', 'ar', 'rw1', 'rw2', 'besag', or 'I')
model = "exchangeable",
##:ARGUMENT: order Defines the \code{order} of the model: for model \code{ar} this defines the order p, in AR(p). Not used for other models at the time being.
order = NULL,
##:ARGUMENT: cyclic Make the group model cyclic? (Only applies to models 'ar1', 'rw1' and 'rw2')
cyclic = FALSE,
##:ARGUMENT: graph The graph spesification (Only applies to model 'besag')
graph = NULL,
##:ARGUMENT: scale.model Scale the intrinsic model (RW1, RW2, BESAG) so the generalized variance is 1. (Default \code{inla.getOption("scale.model.default")}.)
scale.model = NULL,
##:ARGUMENT: adjust.for.con.comp Adjust for connected components when \code{scale.model=TRUE}?
adjust.for.con.comp = TRUE,
##:ARGUMENT: hyper Definition of the hyperparameter(s)
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value for the group correlation or precision in the internal scale.
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if the group correction or precision is assumed to be fixed or random.
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution for the group correlation or precision in the internal scale
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) Prior parameters
param = NULL)
##:SEEALSO: inla
}
`inla.set.control.mix.default` =
function(...)
{
##:EXTRA: The \code{control.mix} -list is set within the corresponding \code{control.family}-list a the mixture of the likelihood is likelihood spesific. (This option is EXPERIMENTAL.)
##:NAME: control.mix
list(
##:ARGUMENT: model The model for the random effect. Currently, only \code{model='gaussian'} is implemented
model = NULL,
##:ARGUMENT: hyper Definition of the hyperparameter(s) for the random effect model chosen
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value(s) for the hyperparameter(s)
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if hyperparmater(s) is/are fixed or random
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for the hyperparmater(s)
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution(s) for the hyperparmater(s)
param = NULL)
##:SEEALSO: inla
}
`inla.set.control.link.default` =
function(...)
{
##:EXTRA: The \code{control.link}-list is set within the corresponding \code{control.family}-list as the link is likelihood-familiy spesific.
##:NAME: control.link
list(
##:ARGUMENT: model The name of the link function/model
model = "default",
##:ARGUMENT: order The \code{order} of the link function, where the interpretation of \code{order} is model-dependent.
order = NULL,
##:ARGUMENT: nq Number of quadrature-points used to do the numerical integration
nq = 15,
##:ARGUMENT: hyper Definition of the hyperparameter(s) for the link model chosen
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value(s) for the hyperparameter(s)
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if hyperparmater(s) is/are fixed or random
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for the hyperparmater(s)
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution(s) for the hyperparmater(s)
param = NULL)
##:SEEALSO: inla
}
`inla.set.f.default` =
function(...)
{
list(diagonal = .Machine$double.eps^0.3833) ## almost 1e-6 on my computer
}
`inla.set.control.expert.default` =
function(...)
{
##:EXTRA:
##:NAME: control.expert
list(
##:ARGUMENT: cpo.manual A boolean variable to decide if the inla-program is to be runned in a manual-cpo-mode. (EXPERT OPTION: DO NOT USE)
cpo.manual = FALSE,
##:ARGUMENT: cpo.idx The index of the data point to remove. (EXPERT OPTION: DO NOT USE)
cpo.idx = -1,
##:ARGUMENT: jp.func The R-function which returns the joint prior, to be defined in \code{jp.Rfile}
jp.func = NULL,
##:ARGUMENT: jp.Rfile The R-file to be sourced to set up a joint prior for the hyperparmaters to be evaluated by \code{jp.func}
jp.Rfile = NULL,
##:ARGUMENT: disable.gaussian.check Disable the check for fast computations with a Gaussian likelihood and identity link
disable.gaussian.check = FALSE
)
##:SEEALSO: inla
}
`inla.set.control.compute.default`=
function(...)
{
##:EXTRA:
##:NAME: control.compute
list(
##:ARGUMENT: openmp.strategy The computational strategy to use: 'small', 'medium', 'large', 'huge' and 'default'. The difference is how the parallelisation is done, and is tuned for 'small'-sized models, 'medium'-sized models, etc. The default option tries to make an educated guess, but this allows to overide this selection. Default is 'default'
openmp.strategy = "default", ## "small", "medium", "large", "huge"
##:ARGUMENT: hyperpar A boolean variable if the marginal for the hyperparameters should be computed. Default TRUE.
hyperpar=TRUE,
##:ARGUMENT: return.marginals A boolean variable if the marginals for the latent field should be returned (although it is computed). Default TRUE
return.marginals=TRUE,
##:ARGUMENT: dic A boolean variable if the DIC-value should be computed. Default FALSE.
dic=FALSE,
##:ARGUMENT: mlik A boolean variable if the marginal likelihood should be computed. Default FALSE.
mlik=TRUE,
##:ARGUMENT: cpo A boolean variable if the cross-validated predictive measures (cpo, pit) should be computed
cpo=FALSE,
##:ARGUMENT: po A boolean variable if the predictive ordinate should be computed
po=FALSE,
##:ARGUMENT: waic A boolean variable if the Watanabe-Akaike information criteria should be computed
waic=FALSE,
##:ARGUMENT: q A boolean variable if binary images of the precision matrix, the reordered precision matrix and the Cholesky triangle should be generated. (Default FALSE.)
q=FALSE,
##:ARGUMENT: config A boolean variable if the internal GMRF approximations be stored. (Default FALSE. EXPERIMENTAL)
config=FALSE,
##:ARGUMENT: smtp The sparse-matrix solver, one of 'smtp' (default) or 'band'
smtp = NULL,
##:ARGUMENT: graph A boolean variable if the graph itself should be returned. (Default FALSE.)
graph = FALSE,
##:ARGUMENT: gdensity A boolean variable if the Gaussian-densities itself should be returned. (Default FALSE.)
gdensity = FALSE)
##:SEEALSO: inla
}
`inla.set.control.family.default`=
function(...)
{
##:EXTRA:
##:NAME: control.family
list(
##:ARGUMENT: hyper Definition of the hyperparameters
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) Initial value for the hyperparameter(s) of the likelihood in the internal scale.
initial=NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for othe hyperparameter(s).
prior=NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution
param=NULL,
##:ARGUMENT: fixed (OBSOLETE!) Boolean variable(s) to say if the hyperparameter(s) is fixed or random.
fixed=NULL,
##:ARGUMENT: link (OBSOLETE! Use \code{control.link=list(model=)} instead.) The link function to use.
link= "default",
##:ARGUMENT: alpha The parameter 'alpha' for the asymmetric Laplace likelihood (default 0.5)
alpha=0.5,
##:ARGUMENT: epsilon The parameter 'epsilon' for the asymmetric Laplace likelihood (default 0.01)
epsilon = 0.01,
##:ARGUMENT: gamma The parameter 'gamma' for the asymmetric Laplace likelihood (default 1.0)
gamma = 1.0,
##:ARGUMENT: sn.shape.max Maximum value for the shape-parameter for Skew Normal observations
sn.shape.max = 5.0,
##:ARGUMENT: gev.scale.xi The internal scaling of the shape-parameter for the GEV distribution. (default 0.01)
gev.scale.xi = 0.01,
##:ARGUMENT: variant This variable is used to give options for various variants of the likelihood, like chosing different parameterisations for example. See the relevant likelihood documentations for options (does only apply to some likelihoods).
variant = 0L,
##:ARGUMENT: control.mix See \code{?control.mix}
control.mix = NULL,
##:ARGUMENT: control.link See \code{?control.link}
control.link = NULL
)
##:SEEALSO: inla
}
`inla.set.control.fixed.default`=
function(...)
{
##:EXTRA:
##:NAME: control.fixed
list(
##:ARGUMENT: expand.factor.strategy The strategy used to expand factors into fixed effects based on their levels. The default strategy is us use the \code{model.matrix}-function for which NA's are not allowed (\code{expand.factor.strategy="model.matrix"}) and levels are possible removed. The alternative option (\code{expand.factor.strategy="inla"}) use an \code{inla}-spesific expansion which expand a factor into one fixed effects for each level, do allow for NA's and all levels are present in the model. In this case, factors MUST BE factors in the data.frame/list and NOT added as \code{...+factor(x1)+...} in the formula only.
expand.factor.strategy = "model.matrix",
##:ARGUMENT: cdf A list of values to compute the CDF for, for all fixed effects
cdf=NULL,
##:ARGUMENT: quantiles A list of quantiles to compute for all fixed effects
quantiles = NULL,
##:ARGUMENT: mean Prior mean for all fixed effects except the intercept. Alternatively, a named list with specific means where name=default applies to unmatched names. For example \code{control.fixed=list(mean=list(a=1, b=2, default=0))} assign 'mean=1' to fixed effect 'a' , 'mean=2' to effect 'b' and 'mean=0' to all others.
mean = 0.0,
##:ARGUMENT: mean.intercept Prior mean for the intercept
mean.intercept = 0.0,
##:ARGUMENT: prec Default precision for all fixed effects except the intercept. Alternatively, a named list with specific means where name=default applies to unmatched names. For example \code{control.fixed=list(prec=list(a=1, b=2, default=0.01))} assign 'prec=1' to fixed effect 'a' , 'prec=2' to effect 'b' and 'prec=0.01' to all others.
prec= 0.001,
##:ARGUMENT: prec.intercept Default precision the intercept (default 0.0)
prec.intercept = 0.0,
##:ARGUMENT: compute Compute marginals for the fixed effects ? (default TRUE)
compute = TRUE,
##:ARGUMENT: correlation.matrix Compute the posterior correlation matrix for all fixed effects? (default FALSE) OOPS: This option will set up appropriate linear combinations and the results are shown as the posterior correlation matrix of the linear combinations. This option will imply \code{control.inla=list(lincomb.derived.correlation.matrix=TRUE)}.
correlation.matrix = FALSE)
##:SEEALSO: inla
}
`inla.set.control.inla.default`=
function(...)
{
family = "gaussian"
xx = list(...)[1]
if (!is.null(xx$family)) {
family = xx$family
}
##:EXTRA:
##:NAME: control.inla
ans = list(
##:ARGUMENT: strategy The strategy to use for the approximations; one of 'gaussian', 'simplified.laplace' (default) or 'laplace'
strategy="simplified.laplace",
##:ARGUMENT: int.strategy The integration strategy to use; one of 'ccd' (default), 'grid' or 'eb' (empirical bayes)
int.strategy="ccd",
##:ARGUMENT: interpolator The interpolator used to compute the marginals for the hyperparameters. One of 'auto', 'nearest', 'quadratic', 'weighted.distance', 'ccd', 'ccdintegrate', 'gridsum', 'gaussian'. Default is 'auto'.
interpolator="auto",
##:ARGUMENT: fast Fast mode? If on, then replace conditional modes in the Laplace approximation with conditional expectation (default TRUE)
fast = TRUE,
##:ARGUMENT: linear.correction Default TRUE for the 'strategy = laplace' option.
linear.correction=NULL,
##:ARGUMENT: h The step-length for the gradient calculations for the hyperparameters. Default 0.01.
h=0.01,
##:ARGUMENT: dz The step-length in the standarised scale for the integration of the hyperparameters. Default 1.0.
dz=1.0,
##:ARGUMENT: diff.logdens The difference of the log.density for the hyperpameters to stop numerical integration using int.strategy='grid'. Default 2.5.
diff.logdens=2.5,
##:ARGUMENT: print.joint.hyper If TRUE, the store also the joint distribution of the hyperparameters (without any costs). Default TRUE.
print.joint.hyper=TRUE,
##:ARGUMENT: force.diagonal A boolean variable, if TRUE, then force the Hessian to be diagonal. (Default FALSE.)
force.diagonal=FALSE,
##:ARGUMENT: skip.configurations A boolean variable; skip configurations if the values at the main axis are to small. (Default TRUE.)
skip.configurations=TRUE,
##:ARGUMENT: mode.known A boolean variable: If TRUE then no optimisation is done. (Default FALSE.)
mode.known=FALSE,
##:ARGUMENT: adjust.weights A boolean variable; If TRUE then just more accurate integration weights. (Default TRUE.)
adjust.weights=TRUE,
##:ARGUMENT: tolerance The tolerance for the optimisation of the hyperparameters. If set, this is the default value for for 'tolerance.f^(2/3)', 'tolerance.g' and 'tolerance.x'; see below.
tolerance = 0.005,
##:ARGUMENT: tolerance.f The tolerance for the absolute change in the log posterior in the optimisation of the hyperparameters.
tolerance.f = NULL,
##:ARGUMENT: tolerance.g The tolerance for the absolute change in the gradient of the log posterior in the optimisation of the hyperparameters.
tolerance.g = NULL,
##:ARGUMENT: tolerance.x The tolerance for the change in the hyperparameters (root-mean-square) in the optimisation of the hyperparameters.
tolerance.x = NULL,
##:ARGUMENT: restart To improve the optimisation, the optimiser is restarted at the found optimum 'restart' number of times.
restart = 0L,
##:ARGUMENT: optimiser The optimiser to use; one of 'gsl', 'domin' or 'default'.
optimiser = "default",
##:ARGUMENT: verbose A boolean variable; run in verbose mode? (Default FALSE)
verbose = NULL,
##:ARGUMENT: reordering Type of reordering to use. (EXPERT OPTION; one of "AUTO", "DEFAULT", "IDENTITY", "REVERSEIDENTITY", "BAND", "METIS", "GENMMD", "AMD", "MD", "MMD", "AMDBAR", "AMDC", "AMDBARC", or the output from \code{inla.qreordering}.)
reordering = "auto",
##:ARGUMENT: cpo.diff Threshold to define when the cpo-calculations are inaccurate. (EXPERT OPTION.)
cpo.diff = NULL,
##:ARGUMENT: npoints Number of points to use in the 'stratey=laplace' approximation
npoints = 9,
##:ARGUMENT: cutoff The cutoff used in the 'stratey=laplace' approximation. (Smaller value is more accurate and more slow.)
cutoff = 1e-4,
##:ARGUMENT: adapt.hessian.mode A boolean variable; should optimisation be continued if the Hessian estimate is void? (Default TRUE)
adapt.hessian.mode = NULL,
##:ARGUMENT: adapt.hessian.max.trials Number of steps in the adaptive Hessian optimisation
adapt.hessian.max.trials = NULL,
##:ARGUMENT: adapt.hessian.scale The scaling of the 'h' after each trial.
adapt.hessian.scale = NULL,
##:ARGUMENT: huge A boolean variable; if TRUE then try to do some of the internal parallisations differently. Hopefully this will be of benefite for 'HUGE' models. (Default FALSE.) [THIS OPTION IS OBSOLETE AND NOT USED!]
huge = FALSE,
##:ARGUMENT: step.len The step-length used to compute numerical derivaties of the log-likelihood
step.len = .Machine$double.eps^(1.0/5.5),
##:ARGUMENT: stencil Number of points in the stencil used to compute the numerical derivaties of the log-likelihood (3, 5 or 7).
stencil = 5L,
##:ARGUMENT: lincomb.derived.only A boolean variable: if TRUE the only compute the marginals for the derived linear combinations and if FALSE, the and also the linear combinations to the graph (Default TRUE)
lincomb.derived.only = TRUE,
##:ARGUMENT: lincomb.derived.correlation.matrix A boolean variable: if TRUE compute also the correlations for the derived linear combinations, if FALSE do not (Default FALSE)
lincomb.derived.correlation.matrix = FALSE,
##:ARGUMENT: diagonal Expert use only! Add a this value on the diagonal of the joint precision matrix.
diagonal = 0.0,
##:ARGUMENT: numint.maxfeval Maximum number of function evaluations in the the numerical integration for the hyperparameters. (Default 10000.)
numint.maxfeval = 100000,
##:ARGUMENT: numint.relerr Relative error requirement in the the numerical integration for the hyperparameters. (Default 1e-5)
numint.relerr = 1e-5,
##:ARGUMENT: numint.abserr Absolute error requirement in the the numerical integration for the hyperparameters. (Default 1e-6)
numint.abserr = 1e-6,
##:ARGUMENT: cmin The minimum value for the negative Hessian from the likelihood. Increasing this value will stabalise the optimisation. (Default 0.0)
cmin = 0.0,
##:ARGUMENT: step.factor The step factor in the Newton-Raphson algorithm saying how large step to take (Default 1.0)
## YES! setting this to a negative values means = 1, EXCEPT the first time (for each thread) where |step.factor| is used.
## This is an hidden option.
step.factor = -0.1,
##:ARGUMENT: global.node.factor The factor which defines the degree required (how many neighbors), as a fraction of n-1, that is required to be classified as a global node and numbered last (whatever the reordering routine says). Here, n, is the size of the graph. (Disabled if larger than 1.)
global.node.factor = 2.0,
##:ARGUMENT: global.node.degree The degree required (number of neighbors) to be classified as a global node and numbered last (whatever the reordering routine says).
global.node.degree = .Machine$integer.max,
##:ARGUMENT: stupid.search Enable or disable the stupid-search-algorithm, if the Hessian calculations reveals that the mode is not found. (Default \code{TRUE}.)
stupid.search = TRUE,
##:ARGUMENT: stupid.search.max.iter Maximum number of iterations allowed for the stupid-search-algorithm.
stupid.search.max.iter = 1000L,
##:ARGUMENT: stupid.search.factor Factor (>=1) to increase the step-length with after each new interation.
stupid.search.factor = 1.05,
##:ARGUMENT: correct Add correction for the Laplace approximation.
correct = FALSE,
##:ARGUMENT: correct.factor Factor used in adjusting the correction factor (default=10) if correct=TRUE
correct.factor = 10.0,
##:ARGUMENT: correct.strategy The strategy used to compute the correction; one of 'simplified.laplace' (default) or 'laplace'
correct.strategy = "simplified.laplace",
##:ARGUMENT: correct.verbose Be verbose when computing the correction?
correct.verbose = FALSE)
## use default Gaussian strategy if the observations are gaussian
if (all(tolower(family) %in% "gaussian"))
ans$strategy = "gaussian"
##:SEEALSO: inla
return (ans)
}
`inla.set.control.predictor.default`=
function(...)
{
##:EXTRA:
##:NAME: control.predictor
list(
##:ARGUMENT: hyper Definition of the hyperparameters.
hyper = NULL,
##:ARGUMENT: fixed (OBSOLETE!) If the precision for the artificial noise is fixed or not (defualt TRUE)
fixed=NULL,
##:ARGUMENT: prior (OBSOLETE!) The prior for the artificial noise
prior=NULL,
##:ARGUMENT: param (OBSOLETE!) Prior parameters for the artificial noise
param=NULL,
##:ARGUMENT: initial (OBSOLETE!) The value of the log precision of the artificial noise
initial=NULL,
##:ARGUMENT: compute A boolean variable; should the marginals for the linear predictor be computed? (Default FALSE.)
compute=FALSE,
##:ARGUMENT: cdf A list of values to compute the CDF for the linear predictor
cdf=NULL,
##:ARGUMENT: quantiles A list of quantiles to compute for the linear predictor
quantiles = NULL,
##:ARGUMENT: cross Cross-sum-to-zero constraints with the linear predictor. All linear predictors with the same level of 'cross' are constrained to have sum zero. Use 'NA' for no contribution. 'Cross' has the same length as the linear predictor (including the 'A' matrix extention). (THIS IS AN EXPERIMENTAL OPTION, CHANGES MAY APPEAR.)
cross=NULL,
##:ARGUMENT: A The observation matrix (matrix or Matrix::sparseMatrix) or a filename with format `i j value'.
A = NULL,
##:ARGUMENT: precision The precision for eta* - A*eta,
precision = exp(15),
##:ARGUMENT: link Define the family-connection for unobserved observations (\code{NA}). \code{link} is integer values which defines the family connection; \code{family[link[idx]]} unless \code{is.na(link[idx])} for which the identity-link is used. The \code{link}-argument only influence the \code{fitted.values} in the \code{result}-object. If \code{is.null(link)} (default) then the identity-link is used for all missing observations. If the length of \code{link} is 1, then this value is replicated with the length of the responce vector. If an element of the responce vector is \code{!NA} then the corresponding entry in \code{link} is not used (but must still be a legal value). Setting this variable implies \code{compute=TRUE}.
link = NULL)
##:SEEALSO: inla
}
`inla.set.control.results.default`=
function(...)
{
##:EXTRA:
##:NAME: control.results
list(
##:ARGUMENT: return.marginals.random A boolean variable; read the marginals for the fterms? (Default TRUE)
return.marginals.random=TRUE,
##:ARGUMENT: return.marginals.predictor A boolean variable; read the marginals for the linear predictor? (Default TRUE)
return.marginals.predictor=TRUE)
##:SEEALSO: inla
}
`inla.set.control.mode.default`=
function(...)
{
## this is internal use only...
##:EXTRA:
##:NAME: control.mode
list(
##:ARGUMENT: result Prevous result from inla(). Use the theta- and x-mode from this run.
result = NULL,
##:ARGUMENT: theta The theta-mode/initial values for theta. This option has preference over result$mode$theta.
theta = NULL,
##:ARGUMENT: x The x-mode/intitial values for x. This option has preference over result$mode$x.
x = NULL,
##:ARGUMENT: restart A boolean variable; should we restart the optimisation from this configuration or fix the mode at this configuration? (Default FALSE.)
restart = FALSE,
##:ARGUMENT: fixed A boolean variable. If TRUE then treat all thetas as known and fixed, and if FALSE then treat all thetas as unknown and random (default).
fixed = FALSE)
##:SEEALSO: inla
}
`inla.set.control.hazard.default` =
function(...)
{
##:EXTRA:
##:NAME: control.hazard
list(
##:ARGUMENT: model The model for the baseline hazard model. One of 'rw1' or 'rw2'. (Default 'rw1'.)
model = "rw1",
##:ARGUMENT: hyper The definition of the hyperparameters.
hyper = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable; is the precision for 'model' fixed? (Default FALSE.)
fixed = FALSE,
##:ARGUMENT: initial (OBSOLETE!) The initial value for the precision.
initial = NULL,
##:ARGUMENT: prior (OBSOLETE!) The prior distribution for the precision for 'model'
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters in the prior distribution
param = NULL,
##:ARGUMENT: constr A boolean variable; shall the 'model' be constrained to sum to zero?
constr = TRUE,
##:ARGUMENT: n.intervals Number of intervals in the baseline hazard. (Default 15)
n.intervals = 15,
##:ARGUMENT: cutpoints The cutpoints to use. If not specified the they are compute from 'n.intervals' and the maximum length of the interval. (Default NULL)
cutpoints = NULL,
##:ARGUMENT: strata.name The name of the stratefication variable for the baseline hazard in the data.frame
strata.name = NULL,
##:ARGUMENT: scale.model Scale the baseline hazard model (RW1, RW2) so the generalized variance is 1. (Default \code{inla.getOption("scale.model.default")}.)
scale.model = NULL)
##:SEEALSO: inla
}
## check control-arguments
`inla.check.control` = function(contr, data = NULL)
{
## This function will signal an error if the arguments in CONTR
## does not match the ones in the corresponding
## `inla.set.XX.default()' routine. EXAMPLE: contr is
## `control.inla' and default arguments is found in
## `inla.set.control.inla.default()'
## Will expand unexpanded names from the names in 'data' first
contr = local({
name = paste("inla.tmp.env", as.character(runif(1)), sep="")
attach(data, name = name, warn.conflicts = FALSE)
ccontr = contr
detach(name, character.only = TRUE)
ccontr
})
stopifnot(!missing(contr))
stopifnot(is.list(contr))
if (length(contr) == 0) {
return(contr)
}
nm = paste(sys.call()[2])
f = paste("inla.set.", nm, ".default()", sep="")
elms = names(inla.eval(f))
if (is.null(names(contr))) {
stop(inla.paste(c("Named elements in in control-argument `", nm, "', is required: ", contr,
"\n\n Valid ones are:\n\t",
inla.paste(sort(elms), sep="\n\t")), sep=""))
}
for(elm in names(contr)) {
if (!is.element(elm, elms)) {
stop(inla.paste(c("Name `", elm,"' in control-argument `", nm, "', is void.\n\n Valid ones are:\n\t",
inla.paste(sort(elms), sep="\n\t")), sep=""))
}
}
return(contr)
}
## test-implementation
##`control.lincomb` = function(precision, verbose)
##{
## aa = match.call()[-1]
## ret = list()
## for(a in names(aa)) {
## if (!missing(a)) {
## xx = get(a)
## names(xx) = a
## ret = c(ret, xx)
## }
## }
## return (ret)
##}
inla.make.completion.function = function(...)
{
my.eval = function(command, envir = parent.frame(),
enclos = if (is.list(envir) || is.pairlist(envir)) parent.frame() else baseenv())
{
return(eval(parse(text = command), envir, enclos))
}
xx = sort(list(...)[[1L]])
my.eval(paste("function(", paste(xx, sep="", collapse=" ,"), ") {
aa = match.call()[-1L]
ret = list()
for(a in names(aa)) {
if (!missing(a)) {
xx = get(a)
names(xx) = a
ret = c(ret, xx)
}
}
return (ret)
}"))
}
control.update = inla.make.completion.function(names(inla.set.control.update.default()))
control.lincomb = inla.make.completion.function(names(inla.set.control.lincomb.default()))
control.group = inla.make.completion.function(names(inla.set.control.group.default()))
control.mix = inla.make.completion.function(names(inla.set.control.mix.default()))
control.link = inla.make.completion.function(names(inla.set.control.link.default()))
control.expert = inla.make.completion.function(names(inla.set.control.expert.default()))
control.compute = inla.make.completion.function(names(inla.set.control.compute.default()))
control.family = inla.make.completion.function(names(inla.set.control.family.default()))
control.fixed = inla.make.completion.function(names(inla.set.control.fixed.default()))
control.inla = inla.make.completion.function(names(inla.set.control.inla.default()))
control.predictor = inla.make.completion.function(names(inla.set.control.predictor.default()))
control.results = inla.make.completion.function(names(inla.set.control.results.default()))
control.mode = inla.make.completion.function(names(inla.set.control.mode.default()))
control.hazard = inla.make.completion.function(names(inla.set.control.hazard.default()))
andrewzm/INLA documentation built on May 10, 2019, 11:12 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
## Export: inla.set.control.lincomb.default
## Export: inla.set.control.update.default
## Export: inla.set.control.group.default
## Export: inla.set.control.mix.default
## Export: inla.set.control.link.default
## Export: inla.set.control.expert.default
## Export: inla.set.control.compute.default
## Export: inla.set.control.family.default
## Export: inla.set.control.fixed.default
## Export: inla.set.control.inla.default
## Export: inla.set.control.predictor.default
## Export: inla.set.control.results.default
## Export: inla.set.control.mode.default
## Export: inla.set.control.hazard.default
## Export: control.lincomb
## Export: control.update
## Export: control.group
## Export: control.mix
## Export: control.link
## Export: control.expert
## Export: control.compute
## Export: control.family
## Export: control.fixed
## Export: control.inla
## Export: control.predictor
## Export: control.results
## Export: control.mode
## Export: control.hazard
### Defines default arguments
`inla.set.control.update.default` =
function(...)
{
##:EXTRA:
##:NAME: control.update
list(
##:ARGUMENT: result Update the joint posterior for the hyperparameters from result
result = NULL
)
##:SEEALSO: inla
}
`inla.set.control.lincomb.default` =
function(...)
{
##:EXTRA:
##:NAME: control.lincomb
list(
##:ARGUMENT: precision The precision for the artificial tiny noise. Default 1e09.
precision = 10^9,
##:ARGUMENT: verbose Use verbose mode for linear combinations if verbose model is set globally. (Default TRUE)
verbose = TRUE)
##:SEEALSO: inla
}
`inla.set.control.group.default` =
function(...)
{
##:EXTRA:
##:NAME: control.group
list(
##:ARGUMENT: model Group model (one of 'exchangable', 'ar1', 'ar', 'rw1', 'rw2', 'besag', or 'I')
model = "exchangeable",
##:ARGUMENT: order Defines the \code{order} of the model: for model \code{ar} this defines the order p, in AR(p). Not used for other models at the time being.
order = NULL,
##:ARGUMENT: cyclic Make the group model cyclic? (Only applies to models 'ar1', 'rw1' and 'rw2')
cyclic = FALSE,
##:ARGUMENT: graph The graph spesification (Only applies to model 'besag')
graph = NULL,
##:ARGUMENT: scale.model Scale the intrinsic model (RW1, RW2, BESAG) so the generalized variance is 1. (Default \code{inla.getOption("scale.model.default")}.)
scale.model = NULL,
##:ARGUMENT: adjust.for.con.comp Adjust for connected components when \code{scale.model=TRUE}?
adjust.for.con.comp = TRUE,
##:ARGUMENT: hyper Definition of the hyperparameter(s)
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value for the group correlation or precision in the internal scale.
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if the group correction or precision is assumed to be fixed or random.
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution for the group correlation or precision in the internal scale
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) Prior parameters
param = NULL)
##:SEEALSO: inla
}
`inla.set.control.mix.default` =
function(...)
{
##:EXTRA: The \code{control.mix} -list is set within the corresponding \code{control.family}-list a the mixture of the likelihood is likelihood spesific. (This option is EXPERIMENTAL.)
##:NAME: control.mix
list(
##:ARGUMENT: model The model for the random effect. Currently, only \code{model='gaussian'} is implemented
model = NULL,
##:ARGUMENT: hyper Definition of the hyperparameter(s) for the random effect model chosen
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value(s) for the hyperparameter(s)
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if hyperparmater(s) is/are fixed or random
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for the hyperparmater(s)
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution(s) for the hyperparmater(s)
param = NULL)
##:SEEALSO: inla
}
`inla.set.control.link.default` =
function(...)
{
##:EXTRA: The \code{control.link}-list is set within the corresponding \code{control.family}-list as the link is likelihood-familiy spesific.
##:NAME: control.link
list(
##:ARGUMENT: model The name of the link function/model
model = "default",
##:ARGUMENT: order The \code{order} of the link function, where the interpretation of \code{order} is model-dependent.
order = NULL,
##:ARGUMENT: nq Number of quadrature-points used to do the numerical integration
nq = 15,
##:ARGUMENT: hyper Definition of the hyperparameter(s) for the link model chosen
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) The initial value(s) for the hyperparameter(s)
initial = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable if hyperparmater(s) is/are fixed or random
fixed = NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for the hyperparmater(s)
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution(s) for the hyperparmater(s)
param = NULL)
##:SEEALSO: inla
}
`inla.set.f.default` =
function(...)
{
list(diagonal = .Machine$double.eps^0.3833) ## almost 1e-6 on my computer
}
`inla.set.control.expert.default` =
function(...)
{
##:EXTRA:
##:NAME: control.expert
list(
##:ARGUMENT: cpo.manual A boolean variable to decide if the inla-program is to be runned in a manual-cpo-mode. (EXPERT OPTION: DO NOT USE)
cpo.manual = FALSE,
##:ARGUMENT: cpo.idx The index of the data point to remove. (EXPERT OPTION: DO NOT USE)
cpo.idx = -1,
##:ARGUMENT: jp.func The R-function which returns the joint prior, to be defined in \code{jp.Rfile}
jp.func = NULL,
##:ARGUMENT: jp.Rfile The R-file to be sourced to set up a joint prior for the hyperparmaters to be evaluated by \code{jp.func}
jp.Rfile = NULL,
##:ARGUMENT: disable.gaussian.check Disable the check for fast computations with a Gaussian likelihood and identity link
disable.gaussian.check = FALSE
)
##:SEEALSO: inla
}
`inla.set.control.compute.default`=
function(...)
{
##:EXTRA:
##:NAME: control.compute
list(
##:ARGUMENT: openmp.strategy The computational strategy to use: 'small', 'medium', 'large', 'huge' and 'default'. The difference is how the parallelisation is done, and is tuned for 'small'-sized models, 'medium'-sized models, etc. The default option tries to make an educated guess, but this allows to overide this selection. Default is 'default'
openmp.strategy = "default", ## "small", "medium", "large", "huge"
##:ARGUMENT: hyperpar A boolean variable if the marginal for the hyperparameters should be computed. Default TRUE.
hyperpar=TRUE,
##:ARGUMENT: return.marginals A boolean variable if the marginals for the latent field should be returned (although it is computed). Default TRUE
return.marginals=TRUE,
##:ARGUMENT: dic A boolean variable if the DIC-value should be computed. Default FALSE.
dic=FALSE,
##:ARGUMENT: mlik A boolean variable if the marginal likelihood should be computed. Default FALSE.
mlik=TRUE,
##:ARGUMENT: cpo A boolean variable if the cross-validated predictive measures (cpo, pit) should be computed
cpo=FALSE,
##:ARGUMENT: po A boolean variable if the predictive ordinate should be computed
po=FALSE,
##:ARGUMENT: waic A boolean variable if the Watanabe-Akaike information criteria should be computed
waic=FALSE,
##:ARGUMENT: q A boolean variable if binary images of the precision matrix, the reordered precision matrix and the Cholesky triangle should be generated. (Default FALSE.)
q=FALSE,
##:ARGUMENT: config A boolean variable if the internal GMRF approximations be stored. (Default FALSE. EXPERIMENTAL)
config=FALSE,
##:ARGUMENT: smtp The sparse-matrix solver, one of 'smtp' (default) or 'band'
smtp = NULL,
##:ARGUMENT: graph A boolean variable if the graph itself should be returned. (Default FALSE.)
graph = FALSE,
##:ARGUMENT: gdensity A boolean variable if the Gaussian-densities itself should be returned. (Default FALSE.)
gdensity = FALSE)
##:SEEALSO: inla
}
`inla.set.control.family.default`=
function(...)
{
##:EXTRA:
##:NAME: control.family
list(
##:ARGUMENT: hyper Definition of the hyperparameters
hyper = NULL,
##:ARGUMENT: initial (OBSOLETE!) Initial value for the hyperparameter(s) of the likelihood in the internal scale.
initial=NULL,
##:ARGUMENT: prior (OBSOLETE!) The name of the prior distribution(s) for othe hyperparameter(s).
prior=NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters for the prior distribution
param=NULL,
##:ARGUMENT: fixed (OBSOLETE!) Boolean variable(s) to say if the hyperparameter(s) is fixed or random.
fixed=NULL,
##:ARGUMENT: link (OBSOLETE! Use \code{control.link=list(model=)} instead.) The link function to use.
link= "default",
##:ARGUMENT: alpha The parameter 'alpha' for the asymmetric Laplace likelihood (default 0.5)
alpha=0.5,
##:ARGUMENT: epsilon The parameter 'epsilon' for the asymmetric Laplace likelihood (default 0.01)
epsilon = 0.01,
##:ARGUMENT: gamma The parameter 'gamma' for the asymmetric Laplace likelihood (default 1.0)
gamma = 1.0,
##:ARGUMENT: sn.shape.max Maximum value for the shape-parameter for Skew Normal observations
sn.shape.max = 5.0,
##:ARGUMENT: gev.scale.xi The internal scaling of the shape-parameter for the GEV distribution. (default 0.01)
gev.scale.xi = 0.01,
##:ARGUMENT: variant This variable is used to give options for various variants of the likelihood, like chosing different parameterisations for example. See the relevant likelihood documentations for options (does only apply to some likelihoods).
variant = 0L,
##:ARGUMENT: control.mix See \code{?control.mix}
control.mix = NULL,
##:ARGUMENT: control.link See \code{?control.link}
control.link = NULL
)
##:SEEALSO: inla
}
`inla.set.control.fixed.default`=
function(...)
{
##:EXTRA:
##:NAME: control.fixed
list(
##:ARGUMENT: expand.factor.strategy The strategy used to expand factors into fixed effects based on their levels. The default strategy is us use the \code{model.matrix}-function for which NA's are not allowed (\code{expand.factor.strategy="model.matrix"}) and levels are possible removed. The alternative option (\code{expand.factor.strategy="inla"}) use an \code{inla}-spesific expansion which expand a factor into one fixed effects for each level, do allow for NA's and all levels are present in the model. In this case, factors MUST BE factors in the data.frame/list and NOT added as \code{...+factor(x1)+...} in the formula only.
expand.factor.strategy = "model.matrix",
##:ARGUMENT: cdf A list of values to compute the CDF for, for all fixed effects
cdf=NULL,
##:ARGUMENT: quantiles A list of quantiles to compute for all fixed effects
quantiles = NULL,
##:ARGUMENT: mean Prior mean for all fixed effects except the intercept. Alternatively, a named list with specific means where name=default applies to unmatched names. For example \code{control.fixed=list(mean=list(a=1, b=2, default=0))} assign 'mean=1' to fixed effect 'a' , 'mean=2' to effect 'b' and 'mean=0' to all others.
mean = 0.0,
##:ARGUMENT: mean.intercept Prior mean for the intercept
mean.intercept = 0.0,
##:ARGUMENT: prec Default precision for all fixed effects except the intercept. Alternatively, a named list with specific means where name=default applies to unmatched names. For example \code{control.fixed=list(prec=list(a=1, b=2, default=0.01))} assign 'prec=1' to fixed effect 'a' , 'prec=2' to effect 'b' and 'prec=0.01' to all others.
prec= 0.001,
##:ARGUMENT: prec.intercept Default precision the intercept (default 0.0)
prec.intercept = 0.0,
##:ARGUMENT: compute Compute marginals for the fixed effects ? (default TRUE)
compute = TRUE,
##:ARGUMENT: correlation.matrix Compute the posterior correlation matrix for all fixed effects? (default FALSE) OOPS: This option will set up appropriate linear combinations and the results are shown as the posterior correlation matrix of the linear combinations. This option will imply \code{control.inla=list(lincomb.derived.correlation.matrix=TRUE)}.
correlation.matrix = FALSE)
##:SEEALSO: inla
}
`inla.set.control.inla.default`=
function(...)
{
family = "gaussian"
xx = list(...)[1]
if (!is.null(xx$family)) {
family = xx$family
}
##:EXTRA:
##:NAME: control.inla
ans = list(
##:ARGUMENT: strategy The strategy to use for the approximations; one of 'gaussian', 'simplified.laplace' (default) or 'laplace'
strategy="simplified.laplace",
##:ARGUMENT: int.strategy The integration strategy to use; one of 'ccd' (default), 'grid' or 'eb' (empirical bayes)
int.strategy="ccd",
##:ARGUMENT: interpolator The interpolator used to compute the marginals for the hyperparameters. One of 'auto', 'nearest', 'quadratic', 'weighted.distance', 'ccd', 'ccdintegrate', 'gridsum', 'gaussian'. Default is 'auto'.
interpolator="auto",
##:ARGUMENT: fast Fast mode? If on, then replace conditional modes in the Laplace approximation with conditional expectation (default TRUE)
fast = TRUE,
##:ARGUMENT: linear.correction Default TRUE for the 'strategy = laplace' option.
linear.correction=NULL,
##:ARGUMENT: h The step-length for the gradient calculations for the hyperparameters. Default 0.01.
h=0.01,
##:ARGUMENT: dz The step-length in the standarised scale for the integration of the hyperparameters. Default 1.0.
dz=1.0,
##:ARGUMENT: diff.logdens The difference of the log.density for the hyperpameters to stop numerical integration using int.strategy='grid'. Default 2.5.
diff.logdens=2.5,
##:ARGUMENT: print.joint.hyper If TRUE, the store also the joint distribution of the hyperparameters (without any costs). Default TRUE.
print.joint.hyper=TRUE,
##:ARGUMENT: force.diagonal A boolean variable, if TRUE, then force the Hessian to be diagonal. (Default FALSE.)
force.diagonal=FALSE,
##:ARGUMENT: skip.configurations A boolean variable; skip configurations if the values at the main axis are to small. (Default TRUE.)
skip.configurations=TRUE,
##:ARGUMENT: mode.known A boolean variable: If TRUE then no optimisation is done. (Default FALSE.)
mode.known=FALSE,
##:ARGUMENT: adjust.weights A boolean variable; If TRUE then just more accurate integration weights. (Default TRUE.)
adjust.weights=TRUE,
##:ARGUMENT: tolerance The tolerance for the optimisation of the hyperparameters. If set, this is the default value for for 'tolerance.f^(2/3)', 'tolerance.g' and 'tolerance.x'; see below.
tolerance = 0.005,
##:ARGUMENT: tolerance.f The tolerance for the absolute change in the log posterior in the optimisation of the hyperparameters.
tolerance.f = NULL,
##:ARGUMENT: tolerance.g The tolerance for the absolute change in the gradient of the log posterior in the optimisation of the hyperparameters.
tolerance.g = NULL,
##:ARGUMENT: tolerance.x The tolerance for the change in the hyperparameters (root-mean-square) in the optimisation of the hyperparameters.
tolerance.x = NULL,
##:ARGUMENT: restart To improve the optimisation, the optimiser is restarted at the found optimum 'restart' number of times.
restart = 0L,
##:ARGUMENT: optimiser The optimiser to use; one of 'gsl', 'domin' or 'default'.
optimiser = "default",
##:ARGUMENT: verbose A boolean variable; run in verbose mode? (Default FALSE)
verbose = NULL,
##:ARGUMENT: reordering Type of reordering to use. (EXPERT OPTION; one of "AUTO", "DEFAULT", "IDENTITY", "REVERSEIDENTITY", "BAND", "METIS", "GENMMD", "AMD", "MD", "MMD", "AMDBAR", "AMDC", "AMDBARC", or the output from \code{inla.qreordering}.)
reordering = "auto",
##:ARGUMENT: cpo.diff Threshold to define when the cpo-calculations are inaccurate. (EXPERT OPTION.)
cpo.diff = NULL,
##:ARGUMENT: npoints Number of points to use in the 'stratey=laplace' approximation
npoints = 9,
##:ARGUMENT: cutoff The cutoff used in the 'stratey=laplace' approximation. (Smaller value is more accurate and more slow.)
cutoff = 1e-4,
##:ARGUMENT: adapt.hessian.mode A boolean variable; should optimisation be continued if the Hessian estimate is void? (Default TRUE)
adapt.hessian.mode = NULL,
##:ARGUMENT: adapt.hessian.max.trials Number of steps in the adaptive Hessian optimisation
adapt.hessian.max.trials = NULL,
##:ARGUMENT: adapt.hessian.scale The scaling of the 'h' after each trial.
adapt.hessian.scale = NULL,
##:ARGUMENT: huge A boolean variable; if TRUE then try to do some of the internal parallisations differently. Hopefully this will be of benefite for 'HUGE' models. (Default FALSE.) [THIS OPTION IS OBSOLETE AND NOT USED!]
huge = FALSE,
##:ARGUMENT: step.len The step-length used to compute numerical derivaties of the log-likelihood
step.len = .Machine$double.eps^(1.0/5.5),
##:ARGUMENT: stencil Number of points in the stencil used to compute the numerical derivaties of the log-likelihood (3, 5 or 7).
stencil = 5L,
##:ARGUMENT: lincomb.derived.only A boolean variable: if TRUE the only compute the marginals for the derived linear combinations and if FALSE, the and also the linear combinations to the graph (Default TRUE)
lincomb.derived.only = TRUE,
##:ARGUMENT: lincomb.derived.correlation.matrix A boolean variable: if TRUE compute also the correlations for the derived linear combinations, if FALSE do not (Default FALSE)
lincomb.derived.correlation.matrix = FALSE,
##:ARGUMENT: diagonal Expert use only! Add a this value on the diagonal of the joint precision matrix.
diagonal = 0.0,
##:ARGUMENT: numint.maxfeval Maximum number of function evaluations in the the numerical integration for the hyperparameters. (Default 10000.)
numint.maxfeval = 100000,
##:ARGUMENT: numint.relerr Relative error requirement in the the numerical integration for the hyperparameters. (Default 1e-5)
numint.relerr = 1e-5,
##:ARGUMENT: numint.abserr Absolute error requirement in the the numerical integration for the hyperparameters. (Default 1e-6)
numint.abserr = 1e-6,
##:ARGUMENT: cmin The minimum value for the negative Hessian from the likelihood. Increasing this value will stabalise the optimisation. (Default 0.0)
cmin = 0.0,
##:ARGUMENT: step.factor The step factor in the Newton-Raphson algorithm saying how large step to take (Default 1.0)
## YES! setting this to a negative values means = 1, EXCEPT the first time (for each thread) where |step.factor| is used.
## This is an hidden option.
step.factor = -0.1,
##:ARGUMENT: global.node.factor The factor which defines the degree required (how many neighbors), as a fraction of n-1, that is required to be classified as a global node and numbered last (whatever the reordering routine says). Here, n, is the size of the graph. (Disabled if larger than 1.)
global.node.factor = 2.0,
##:ARGUMENT: global.node.degree The degree required (number of neighbors) to be classified as a global node and numbered last (whatever the reordering routine says).
global.node.degree = .Machine$integer.max,
##:ARGUMENT: stupid.search Enable or disable the stupid-search-algorithm, if the Hessian calculations reveals that the mode is not found. (Default \code{TRUE}.)
stupid.search = TRUE,
##:ARGUMENT: stupid.search.max.iter Maximum number of iterations allowed for the stupid-search-algorithm.
stupid.search.max.iter = 1000L,
##:ARGUMENT: stupid.search.factor Factor (>=1) to increase the step-length with after each new interation.
stupid.search.factor = 1.05,
##:ARGUMENT: correct Add correction for the Laplace approximation.
correct = FALSE,
##:ARGUMENT: correct.factor Factor used in adjusting the correction factor (default=10) if correct=TRUE
correct.factor = 10.0,
##:ARGUMENT: correct.strategy The strategy used to compute the correction; one of 'simplified.laplace' (default) or 'laplace'
correct.strategy = "simplified.laplace",
##:ARGUMENT: correct.verbose Be verbose when computing the correction?
correct.verbose = FALSE)
## use default Gaussian strategy if the observations are gaussian
if (all(tolower(family) %in% "gaussian"))
ans$strategy = "gaussian"
##:SEEALSO: inla
return (ans)
}
`inla.set.control.predictor.default`=
function(...)
{
##:EXTRA:
##:NAME: control.predictor
list(
##:ARGUMENT: hyper Definition of the hyperparameters.
hyper = NULL,
##:ARGUMENT: fixed (OBSOLETE!) If the precision for the artificial noise is fixed or not (defualt TRUE)
fixed=NULL,
##:ARGUMENT: prior (OBSOLETE!) The prior for the artificial noise
prior=NULL,
##:ARGUMENT: param (OBSOLETE!) Prior parameters for the artificial noise
param=NULL,
##:ARGUMENT: initial (OBSOLETE!) The value of the log precision of the artificial noise
initial=NULL,
##:ARGUMENT: compute A boolean variable; should the marginals for the linear predictor be computed? (Default FALSE.)
compute=FALSE,
##:ARGUMENT: cdf A list of values to compute the CDF for the linear predictor
cdf=NULL,
##:ARGUMENT: quantiles A list of quantiles to compute for the linear predictor
quantiles = NULL,
##:ARGUMENT: cross Cross-sum-to-zero constraints with the linear predictor. All linear predictors with the same level of 'cross' are constrained to have sum zero. Use 'NA' for no contribution. 'Cross' has the same length as the linear predictor (including the 'A' matrix extention). (THIS IS AN EXPERIMENTAL OPTION, CHANGES MAY APPEAR.)
cross=NULL,
##:ARGUMENT: A The observation matrix (matrix or Matrix::sparseMatrix) or a filename with format `i j value'.
A = NULL,
##:ARGUMENT: precision The precision for eta* - A*eta,
precision = exp(15),
##:ARGUMENT: link Define the family-connection for unobserved observations (\code{NA}). \code{link} is integer values which defines the family connection; \code{family[link[idx]]} unless \code{is.na(link[idx])} for which the identity-link is used. The \code{link}-argument only influence the \code{fitted.values} in the \code{result}-object. If \code{is.null(link)} (default) then the identity-link is used for all missing observations. If the length of \code{link} is 1, then this value is replicated with the length of the responce vector. If an element of the responce vector is \code{!NA} then the corresponding entry in \code{link} is not used (but must still be a legal value). Setting this variable implies \code{compute=TRUE}.
link = NULL)
##:SEEALSO: inla
}
`inla.set.control.results.default`=
function(...)
{
##:EXTRA:
##:NAME: control.results
list(
##:ARGUMENT: return.marginals.random A boolean variable; read the marginals for the fterms? (Default TRUE)
return.marginals.random=TRUE,
##:ARGUMENT: return.marginals.predictor A boolean variable; read the marginals for the linear predictor? (Default TRUE)
return.marginals.predictor=TRUE)
##:SEEALSO: inla
}
`inla.set.control.mode.default`=
function(...)
{
## this is internal use only...
##:EXTRA:
##:NAME: control.mode
list(
##:ARGUMENT: result Prevous result from inla(). Use the theta- and x-mode from this run.
result = NULL,
##:ARGUMENT: theta The theta-mode/initial values for theta. This option has preference over result$mode$theta.
theta = NULL,
##:ARGUMENT: x The x-mode/intitial values for x. This option has preference over result$mode$x.
x = NULL,
##:ARGUMENT: restart A boolean variable; should we restart the optimisation from this configuration or fix the mode at this configuration? (Default FALSE.)
restart = FALSE,
##:ARGUMENT: fixed A boolean variable. If TRUE then treat all thetas as known and fixed, and if FALSE then treat all thetas as unknown and random (default).
fixed = FALSE)
##:SEEALSO: inla
}
`inla.set.control.hazard.default` =
function(...)
{
##:EXTRA:
##:NAME: control.hazard
list(
##:ARGUMENT: model The model for the baseline hazard model. One of 'rw1' or 'rw2'. (Default 'rw1'.)
model = "rw1",
##:ARGUMENT: hyper The definition of the hyperparameters.
hyper = NULL,
##:ARGUMENT: fixed (OBSOLETE!) A boolean variable; is the precision for 'model' fixed? (Default FALSE.)
fixed = FALSE,
##:ARGUMENT: initial (OBSOLETE!) The initial value for the precision.
initial = NULL,
##:ARGUMENT: prior (OBSOLETE!) The prior distribution for the precision for 'model'
prior = NULL,
##:ARGUMENT: param (OBSOLETE!) The parameters in the prior distribution
param = NULL,
##:ARGUMENT: constr A boolean variable; shall the 'model' be constrained to sum to zero?
constr = TRUE,
##:ARGUMENT: n.intervals Number of intervals in the baseline hazard. (Default 15)
n.intervals = 15,
##:ARGUMENT: cutpoints The cutpoints to use. If not specified the they are compute from 'n.intervals' and the maximum length of the interval. (Default NULL)
cutpoints = NULL,
##:ARGUMENT: strata.name The name of the stratefication variable for the baseline hazard in the data.frame
strata.name = NULL,
##:ARGUMENT: scale.model Scale the baseline hazard model (RW1, RW2) so the generalized variance is 1. (Default \code{inla.getOption("scale.model.default")}.)
scale.model = NULL)
##:SEEALSO: inla
}
## check control-arguments
`inla.check.control` = function(contr, data = NULL)
{
## This function will signal an error if the arguments in CONTR
## does not match the ones in the corresponding
## `inla.set.XX.default()' routine. EXAMPLE: contr is
## `control.inla' and default arguments is found in
## `inla.set.control.inla.default()'
## Will expand unexpanded names from the names in 'data' first
contr = local({
name = paste("inla.tmp.env", as.character(runif(1)), sep="")
attach(data, name = name, warn.conflicts = FALSE)
ccontr = contr
detach(name, character.only = TRUE)
ccontr
})
stopifnot(!missing(contr))
stopifnot(is.list(contr))
if (length(contr) == 0) {
return(contr)
}
nm = paste(sys.call()[2])
f = paste("inla.set.", nm, ".default()", sep="")
elms = names(inla.eval(f))
if (is.null(names(contr))) {
stop(inla.paste(c("Named elements in in control-argument `", nm, "', is required: ", contr,
"\n\n Valid ones are:\n\t",
inla.paste(sort(elms), sep="\n\t")), sep=""))
}
for(elm in names(contr)) {
if (!is.element(elm, elms)) {
stop(inla.paste(c("Name `", elm,"' in control-argument `", nm, "', is void.\n\n Valid ones are:\n\t",
inla.paste(sort(elms), sep="\n\t")), sep=""))
}
}
return(contr)
}
## test-implementation
##`control.lincomb` = function(precision, verbose)
##{
## aa = match.call()[-1]
## ret = list()
## for(a in names(aa)) {
## if (!missing(a)) {
## xx = get(a)
## names(xx) = a
## ret = c(ret, xx)
## }
## }
## return (ret)
##}
inla.make.completion.function = function(...)
{
my.eval = function(command, envir = parent.frame(),
enclos = if (is.list(envir) || is.pairlist(envir)) parent.frame() else baseenv())
{
return(eval(parse(text = command), envir, enclos))
}
xx = sort(list(...)[[1L]])
my.eval(paste("function(", paste(xx, sep="", collapse=" ,"), ") {
aa = match.call()[-1L]
ret = list()
for(a in names(aa)) {
if (!missing(a)) {
xx = get(a)
names(xx) = a
ret = c(ret, xx)
}
}
return (ret)
}"))
}
control.update = inla.make.completion.function(names(inla.set.control.update.default()))
control.lincomb = inla.make.completion.function(names(inla.set.control.lincomb.default()))
control.group = inla.make.completion.function(names(inla.set.control.group.default()))
control.mix = inla.make.completion.function(names(inla.set.control.mix.default()))
control.link = inla.make.completion.function(names(inla.set.control.link.default()))
control.expert = inla.make.completion.function(names(inla.set.control.expert.default()))
control.compute = inla.make.completion.function(names(inla.set.control.compute.default()))
control.family = inla.make.completion.function(names(inla.set.control.family.default()))
control.fixed = inla.make.completion.function(names(inla.set.control.fixed.default()))
control.inla = inla.make.completion.function(names(inla.set.control.inla.default()))
control.predictor = inla.make.completion.function(names(inla.set.control.predictor.default()))
control.results = inla.make.completion.function(names(inla.set.control.results.default()))
control.mode = inla.make.completion.function(names(inla.set.control.mode.default()))
control.hazard = inla.make.completion.function(names(inla.set.control.hazard.default()))
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.