InitExp : Play and visualize EM Algorithm on Gaussian Mixture datasets
This R package is a collection of wrapper functions and test scripts that investigate the property of the Expectation Maximization algorithm. Specifically, it demonstrates some intriguing behavior of EM over a highly non-convex likelihood function.
#Install To install directly from github, open a terminal, type R, then
devtools::install_github('htso/InitExp')
#Dependencies You need my Hext package from github, from a terminal, type
devtools::install_github('htso/Hext')
as well as these packages on CRAN,
install.packages("Hext", "mvtnorm", "corpcor", "ellipse")
#Datasets I include a couple of generic datasets, which are different flavor of gaussian mixture from low to high dimension. To load a dataset, just type
data(simdat2single)
#Run I provide twp demos in the /demo subfolder. To run a demo,
demo("init-demo", package="InitExp")
#WARNING Some of these scripts may take a long time to finish.
#Platforms Tested it on Linux (ubuntu 14.04).