ggmfit: Iterative proportional fitting of graphical Gaussian model

In gRim: Graphical Interaction Models

Iterative proportional fitting of graphical Gaussian model

Description

Fit graphical Gaussian model by iterative proportional fitting.

Usage

ggmfit(
  S,
  n.obs,
  glist,
  start = NULL,
  eps = 1e-12,
  iter = 1000,
  details = 0,
  ...
)

Arguments

S	Empirical covariance matrix
n.obs	Number of observations
glist	Generating class for model (a list)
start	Initial value for concentration matrix
eps	Convergence criterion
iter	Maximum number of iterations
details	Controlling the amount of output.
...	Optional arguments; currently not used

Details

ggmfit is based on a C implementation. ggmfitr is implemented purely in R (and is provided mainly as a benchmark for the C-version).

Value

A list with

lrt	Likelihood ratio statistic (-2logL)
df	Degrees of freedom
logL	log likelihood
K	Estimated concentration matrix (inverse covariance matrix)

Author(s)

Søren Højsgaard, [e-mail]

See Also

cmod, loglin

Examples

## Fitting "butterfly model" to mathmark data
## Notice that the output from the two fitting functions is not
## entirely identical.
data(math)
ddd <- cov.wt(math, method="ML")
glist <- list(c("al", "st", "an"), c("me", "ve", "al"))
ggmfit (ddd$cov, ddd$n.obs, glist)
ggmfitr(ddd$cov, ddd$n.obs, glist)

gRim documentation built on Oct. 2, 2023, 9:06 a.m.