foehnix.control: foehnix Two-Component Mixture-Model Control Object

In retostauffer/Rfoehnix: Objective Foehn Diagnosis

foehnix Two-Component Mixture-Model Control Object

Description

Used to control the foehnix mixture models.

Usage

foehnix.control(
  family,
  switch,
  left = -Inf,
  right = Inf,
  truncated = FALSE,
  standardize = TRUE,
  maxit = 100L,
  tol = 1e-08,
  force.inflate = FALSE,
  glmnet.control = NULL,
  verbose = TRUE,
  ...
)

Arguments

family	character specifying the distribution of the components in the mixture model. Allowed: "gaussian" and "logistic". For experts: custom foehnix.family objects can be provided as well.
switch	logical whether or not the two components should be switched. By default (switch = FALSE) the component which shows higher values of y is assumed to be the foehn cluster! Depending on what your covariate is you might need to switch the clusters (by setting switch = TRUE).
left	default is -Inf, left censoring or truncation point. See also input right and input truncated. Can be set to any finite numeric value.
right	default is Inf, right censoring or truncation point. See also input left and input truncated. Can be set to any finite numeric value.
truncated	logical. If set to TRUE truncation is used instead of censoring. This only affects the foehnix model estimate if input left and/or input right are specified.
standardize	logical flag, default is TRUE. Whether or not the model matrix for the concomitant model should be standardized for model estimation (recommended).
maxit	control argument for the iterative solvers. Default is 100L, the maximum number of iterations for the EM algorithm and the IWLS backfitting algorithm for the concomitant model. If a vector of length two is provided the first value is used for the EM algorithm, the second for the IWLS backfitting.
tol	similar as for maxit. Used to identify convergence of the iterative solvers. Default is 1e-8, if two values are provided the first will be used for the EM algorithm, the second one for the IWLS backfitting procedure. If set to -Inf maxit will be used as stopping criteria.
force.inflate	logical, default is FALSE. foehnix creates a strictly regular time series object by inflating the data set using the smallest time interval in the data set. If the inflation rate is larger than 2 the script will stop except the user forces inflation by specifying force.inflate = TRUE. See 'Details' section for more information.
glmnet.control	an object of class glmnet.control containing the arguments for the glmnet function (experimental).
verbose	logical, if set to FALSE output is suppressed.
...	currently sent to hell.

Details

Inflation: foehnix models are based on time series objects. For some methods (e.g., to create nice and easy to read time series plots and count statistics) foehnix inflates the time series object using the smallest time interval in the data set. This can, possibly, yield very large data sets. Thus, foehnix is pre-calculating the inflation rate, the fraction between the length of the inflated data set versus the length of the data set provided by the user. If this inflation rate exceeds 2 the script will raise an error!

In this case, the user is kindly asked to check if the time series object (input data). A possible scenario: a user is performing foehn diagnosis using 5 years of data from one station with 10 minute observations. This yields (neglecting leap years) 5 * 365 * 144 = 262.800 observations. Imagine that there is one incorrect observation reported one second after one of the regular 10 minute records. The smallest time increment would thus be 1 second. This would yield an inflated time series object with a total record length of 5 * 365 * 86.400 = 157.680.000. Even if only filled with missing values (NA) this will be extremely memory demanding. To avoid this, foehnix will stop in such situations. However, the user is allowed to overrule this condition by setting the force.inflate option to TRUE.

Author(s)

Reto Stauffer

See Also

foehnix, foehnix.family.

retostauffer/Rfoehnix documentation built on June 5, 2023, 11:39 p.m.