Package 'SAGMM'

Return Gamma, a sequence of gain factors

Description

Generate a series of gain factors.

Usage

gainFactors(Number, Burnin)

Arguments

Number	Number of values required.
Burnin	Number of 'Burnin' values at the beginning of sequence.

Value

Gamma, a vector of gain factors.

Examples

g<-gainFactors(10^4, 2*10^3)

---

Generate data for simulations to test the SAGMM package..

Description

This function is primarily a convienence wrapper for MixSim.

Usage

generateSimData(ngroups = 5, Dimensions = 5, Number = 10^4)

Arguments

ngroups	Number of mixture components. Default 5.
Dimensions	number of Dimensions. Default 5.
Number	number of samples. Default 10^4.

Value

List of results: X, Y, simobject.

Examples

sims<-generateSimData(ngroups=10, Dimensions=10, Number=10^4)
sims<-generateSimData()

---

SAGMM: A package for Clustering via Stochastic Approximation and Gaussian Mixture Models.

Description

The SAGMM package allows for computation of gaussian mixture models using stochastic approximation to increase efficiency with large data sets. The primary function SAGMMFit allows this to be performed in a relative flexible manner.

Author(s)

Andrew T. Jones and Hien D. Nguyen

References

Nguyen & Jones (2018). Big Data-Appropriate Clustering via Stochastic Approximation and Gaussian Mixture Models. In Data Analytics (pp. 79-96). CRC Press.

---

Clustering via Stochastic Approximation and Gaussian Mixture Models (GMM)

Description

Fit a GMM via Stochastic Approximation. See Reference.

Usage

SAGMMFit(X, Y = NULL, Burnin = 5, ngroups = 5, kstart = 10,
  plot = FALSE)

Arguments

X	numeric matrix of the data.
Y	Group membership (if known). Where groups are integers in 1:ngroups. If provided ngroups can
Burnin	Ratio of observations to use as a burn in before algorithm begins.
ngroups	Number of mixture components. If Y is provided, and groups is not then is overridden by Y.
kstart	number of kmeans starts to initialise.
plot	If TRUE generates a plot of the clustering.

Value

A list containing

Cluster	The clustering of each observation.
plot	A plot of the clustering (if requested).
l2	Estimate of Lambda^2
ARI1	Adjusted Rand Index 1 - using k-means
ARI2	Adjusted Rand Index 2 - using GMM Clusters
ARI3	Adjusted Rand Index 3 - using intialiation k-means
KM	Initial K-means clustering of the data.
pi	The cluster proportions (vector of length ngroups)
tau	tau matrix of conditional probabilities.
fit	Full output details from inner C++ loop.

Author(s)

Andrew T. Jones and Hien D. Nguyen

References

Nguyen & Jones (2018). Big Data-Appropriate Clustering via Stochastic Approximation and Gaussian Mixture Models. In Data Analytics (pp. 79-96). CRC Press.

Examples

sims<-generateSimData(ngroups=10, Dimensions=10, Number=10^4)
res1<-SAGMMFit(sims$X, sims$Y)
res2<-SAGMMFit(sims$X, ngroups=5)

---