snem: EM algorithm for multivariate skew normal distribution.

In snem: EM Algorithm for Multivariate Skew-Normal Distribution with Overparametrization

Description

EM algorithm in closed form.

Usage

snem(
  x,
  eps = 0.9,
  iter.eps = 10^-6,
  stop.rule = c("parameter", "log-likelihood")
)

Arguments

x	A data matrix. Each row is an observation vector.
eps	Weight parameter with 0 ≤ eps < 1. Default is 0.9.
iter.eps	Convergence threshold. Default is 10^-6.
stop.rule	"parameter": The difference of the parameter is used as a stopping rule. "log-likelihood" The difference of the log-likelihood is used as a stopping rule.

Details

The parameter eps is a tuning parameter which ensures that an initial covariance matrix is positive semi-definite.

Value

Location parameter (mu), covariance matrix (omega), skewness parameter (delta), and another expression of skewness parameter (lambda).

References

Abe, T., Fujisawa, H., and Kawashima, T. (2019) EM algorithm using overparametrization for multivariate skew-normal distribution, in preparation.

Examples

library(sn)
data(ais, package="sn")
x <- ais[c("BMI")]
snem(x, stop.rule ="log-likelihood")

Example output

Loading required package: stats4

Attaching package: ‘sn’

The following object is masked from ‘package:stats’:

    sd


stopping rule:  log-likelihood 
iteration:  64 
log-likelihood:  -490.1021 
mu 
         [,1]
[1,] 19.94748
Omega 
         [,1]
[1,] 17.21203
delta 
          [,1]
[1,] 0.9200908
lambda 
         [,1]
[1,] 2.348938

snem documentation built on March 26, 2020, 5:34 p.m.