fit.rcm: Fit using the EM algorithm

In AEBilgrau/correlateR: Fast correlations and covariances

Description

Fit the RCM using the modified EM algorithm.

Usage

fit.rcm(S, ns, Psi.init, nu.init, method = c("EM", "pooled",
  "approxMLE"), conf.lvl = 0.95, max.ite = 1000, eps = 0.001,
  verbose = FALSE)

Arguments

S	A list of square scatter matrices of the same size.
ns	A vector of group sample sizes corresponding to the scatter matrices in S.
Psi.init	A matrix giving the initial estimate of Psi. Default starting value is the scaled pooled sample covariance matrix.
nu.init	A numeric of length one giving the initial estimate of nu. Default is sum(ns).
method	A character giving the method to be used or abbreviation hereof.
conf.lvl	The confidence level. Default is 0.95.
max.ite	A integer of length one giving the maximum number of iterations allowed. Default is 1000.
eps	The convergence criterion.
verbose	If true, the differences in log-likelihood for each iteration is printed out.

Value

A named list of length 3 with the elements:

Psi	A matrix giving the estimate of Psi.
nu	A number giving the estimate of nu.
iterations	A integer giving the number of iterations used.
loglik	A numeric giving the value of the log-likelihood in the last iteration.

Examples

nss <- c(40, 20, 30, 10)
print(Psii <- drop(rwishart(1)))  # Expected covariance
nuu <- 7
SS <- createRCMData(ns = nss, psi = Psii, nu = nuu)

fit.rcm(SS, nss, method = "EM", verbose = TRUE)
fit.rcm(SS, nss, method = "pool", verbose = TRUE)
fit.rcm(SS, nss, method = "approxMLE", verbose = FALSE)

AEBilgrau/correlateR documentation built on Nov. 15, 2019, 9:21 a.m.