control.fit.default: Default variables in control.fit

In eirikmn/bremla: Bayesian Regression Modeling of Layer-Counted Archives

Default variables in control.fit

Description

Sets the default variables in the list control.fit used to specify the fitting procedure. The list contains the following arguments:

noise

Character which noise process is assumed for the residuals. "iid", "ar1" and "ar2" are currently. However, if synchronization is to be performed only "ar1" is currently supported. Default input is "ar1".

method

Character describing which method should be used to fit the process. Only inla is currently supported, but more will hopefully come in the future.

verbose

Boolean telling whether or not the inla-program should run in verbose mode. Default FALSE.

hyperprior

List corresponding to the hyper object in INLA's f function which specifies the priors of the hyperparameters. Default is NULL which sets INLA's default priors.

improve.fixed

Boolean indicating whether or not the least squares estimate of the fixed values should be used as initial values for the optimization procedure in INLA. This feature has not been extensively tested so it might be unstable. Default FALSE

ncores

Integer used to set the number of cores to be used in the inla fitting procedure. Default value is 1.

transform

Character indicating which transformation should be applied to the observations. Can be "log" or "identity". If synchronization is intended then only "identity" is currently supported, which is also the default value here.

rgeneric

List containing necessary arguments for specifying a custom model for the noise component of the layer increments using the rgeneric framework of INLA. Note that this alternative generally performs less well than using already implemented model components in INLA used when noise is specified. Must include:

from.theta

List of function specifying transformation from internal to original parameter scaling. Function of theta (internal parameters). length(from.theta) must correspond to length(theta). This is only used for transforming the parameters and can not be used in the functions below.

param.names

Vector of names for parameters (optional).

Q

Function that computes the precision matrix of the noise model. Must be a function of theta, n and ntheta, and must return a sparse matrix.

log.prior

Function that computes the logarithm of the prior of the internal parameter theta. Must be a function of theta, n and ntheta (even if they are not used).

Usage

control.fit.default()

Value

Returns a list including default values for all variables in control.fit.

Author(s)

Eirik Myrvoll-Nilsen, [e-mail]

See Also

bremla,bremla_modelfitter,set.options

eirikmn/bremla documentation built on Jan. 25, 2025, 4:41 a.m.