gmm-learn {stochmod}R Documentation

Gaussian Mixture Models

Description

Expectation Maximization algorithm for Gaussian Mixture Models

Usage

GMM.learn( xL, K, vL = NULL, gmm.init = NULL, cov.reg = 0.0, tol =
1e-03, LLstop = Inf, min.iter = 3, max.iter = Inf )

Arguments

xL Either a matrix or a list of matrices containing training observation sequences, with one sample per row
K Desired number of components
vL Either a matrix or a list of matrices containing validation observation sequences, with one sample per row
gmm.init Optional initial model, can be partially specified
cov.reg Covariance matrix regularization (towards identity), value must be in [0, 1]
tol Stopping criterion: relative tolerance on the log-likelihood
LLstop Stopping criterion: hard bound on the log-likelihood value
min.iter At least this number of EM iterations is preformed before validation and tolerance stopping criteria are triggered
max.iter Stoppint criterion: maximum number of iterations

Details

Learns a maximum likelihood GMM given the data

Value

A Gaussian Mixture Model defined by:

mu [K x p] matrix of component means
sigma [K x p x p] array of component covariance matrices
pi [K x 1] vector of mixture coefficients

Author(s)

Artem Sokolov (Artem.Sokolov@gmail.com)


[Package stochmod version 1.2 Index]