README.md

In suren-rathnayake/EMMIXmfa: Mixture Models with Component-Wise Factor Analyzers

EMMIXmfa

This package provides functions for cluster analysis using mixtures of factor analyzers (MFA) and mixture of common factor analyzers (MCFA) models.

MFA and MCFA models belong to the class of finite mixture models, and they adopt component-wise factor analyzers to multivariate data. Component distributions can either be from the family of multivariate normals or from the family of multivariate t-distributions. Maximum likelihood estimators of model parameters are obtained using the Expectation-Maximization algorithm.

Installation

Installingdirectly from using the devtools package.

library(devtools)
install_github("suren-rathnayake/EMMIXmfa")

Usage Example

Fitting a MFA model with there components using two factors for the Iris data available in R can be done using,

# Fit a Gaussian mixture model using MFA
mfa_fit <- mfa(Y = iris[, -5], g = 3, q = 2, sigma_type = "common", D_type = "common")

# Fit a Gaussian mixture model using MCFA
set.seed(1984)
mcfa_fit <- mcfa(Y = iris[, -5], g = 3, q = 2)

The groupings can be visualized in the q-dimensional factor space.

plot_factors(mcfa_fit)

In the case where labeling of the data is available, EMMIXmfa has functionality calculate Adjust Rand Index

ari(mcfa_fit$clust, iris[, 5])
# 0.9602777

and the number of misallocations

minmis(mcfa_fit$clust, iris[, 5])
# 2

Functions mfa and mcfa fits multivariate normals to the data, fitting t-distributions can be achieved using mtfa and mctfa function. Further, there are functions to generate data from a emmix models (rmix), estimate factor scores (factor_scores), estimate adjusted Rand Index (ari), find the number of misallocations (minmis), among others.

suren-rathnayake/EMMIXmfa documentation built on May 17, 2019, 8:45 p.m.