msg: Matrix shrinkage by g-modeling method

In sdzhao/cole: COmpound Loss Emporium

Matrix shrinkage by g-modeling method

Description

The function use shrinkage method to estimate covariance matrix with given multinormal data. This method applies empirical method and approximate the optimal separable decision rule by EM algorithm.

Estimates covariance matrix for data from Gaussian distributino

Usage

msg(
  X,
  K = 10,
  n_start = 20,
  centered = FALSE,
  maxit = 200,
  tol = 1e-04,
  verbose = FALSE
)

msg(
  X,
  K = 10,
  n_start = 20,
  centered = FALSE,
  maxit = 200,
  tol = 1e-04,
  verbose = FALSE
)

Arguments

X	n x p data matrix, each column is a feature
K	number of clusters
centered	whether to centralize each column of X
maxit	maximum number of iterations
tol	error tolerance of convergence of log likelihood
verbose	whether to output the error in each iteration
d	interger. It controls the number of support points. Default value is 10.

Value

• est_cov: p x p matrix of estiamted covariance matrix.

estimated p x p covariance matrix

Examples

## set parameters
p = 10
d = 20
n = 100
## generate multivariate normal data X, with mean of zeros and identity matrix as covariance matrix.
X = matrix(rnorm(n*p,0,1), n, p)
## estimate covariance with given data X
cov = msg(X=X, d=d)


## generate data
n = 100; p = 30
set.seed(1)
sigma.x = matrix(0.9,p,p)
diag(sigma.x) = 1
X = mvrnorm(n,sigma.x)
## loss of msg
norm(sigma.x - msg(X,K=p),'f')

sdzhao/cole documentation built on May 2, 2022, 9:42 a.m.