EMGauss {AdaptGauss}

EM Algorithm for GMM

Description

Expectation-Maximization algorithm to calculate optimal Gaussian Mixture Model for given data in one Dimension.

Usage

EMGauss(Data, K, Means, SDs,Weights, MaxNumberofIterations,fast)

Arguments

Data

vector of data points

K

estimated amount of Gaussian Kernels

Means

vector(1:L), Means of Gaussians, L == Number of Gaussians

SDs

estimated Gaussian Kernels = standard deviations

Weights

optional, relative number of points in Gaussians (prior probabilities): sum(Weights) ==1, default weight is 1/L

MaxNumberofIterations

Optional, Number of Iterations; default=10

fast

Default: FALSE: Using mclust's EM see function densityMclust of that package, TRUE: Naive but faster EM implementation, which may be numerical unstable, because log(gauss) is not used

Details

No adding or removing of Gaussian kernels. Number of Gaussian hast to be set by the length of the vector of Means, SDs and Weights. This EM is only for univariate data. For multivariate data see package mclust

Value

List with

Means

means of GMM generated by EM algorithm

SDs

standard deviations of GMM generated by EM algorithm

Weights

prior probabilities of Gaussians

Author(s)

Onno Hansen-Goos, Michael Thrun, Florian Lerch

References

Bishop, Christopher M. Pattern recognition and machine learning. springer, 2006, p 435 ff

See Also

AdaptGauss