profile.eicm: Likelihood profiles for EICMs

In eicm: Explicit Interaction Community Models

Likelihood profiles for EICMs

Description

Computes the profile (penalized) likelihood for all (or only one) estimated parameters in a EICM model.

Usage

## S3 method for class 'eicm'
profile(
  fitted,
  all.pars = TRUE,
  parmatrix,
  species,
  parameter,
  step = 0.3,
  ncores = parallel::detectCores(),
  alpha = 0.01,
  ...
)

Arguments

fitted	the fitted EICM model.
all.pars	logical. Compute for all model parameters?
parmatrix	if all.pars=FALSE, in which matrix is the parameter of interest, "env" or "sp"?
species	if all.pars=FALSE, in which row of parmatrix is the parameter of interest?
parameter	if all.pars=FALSE, in which column of parmatrix is the parameter of interest?
step	the step increments/decrements at which to compute the likelihood profile points.
ncores	the number of CPU cores to use when computing profiles for all parameters.
alpha	highest significance level that will be guaranteed for this profile.
...	additional argument(s) for methods

Details

Likelihod profiles will use the same regularization settings that were used in model fitting.

Value

The same model object updated with a new profile component.

Note

Confidence intervals should not be computed on a model whose terms have been selected.

This function is optimized for computing profiles of multiple parameters simultaneously (in parallel).

Examples

# load the included parameterized model
data(truemodel)

# realize the model
occurrences <- predict(truemodel, nrepetitions=1)

# fit the model without species interactions
fitted <- eicm(occurrences, n.latent=2, mask.sp=0, do.selection=FALSE)$fitted.model

# compute likelihood profiles for all parameters
fitted <- profile(fitted, ncores=2)

# plot the first 9 profiles
par(mfrow=c(3, 3))
dummy <- lapply(fitted$profile[1:9], plot)

eicm documentation built on May 31, 2023, 5:20 p.m.