Package: autoMFA 1.1.0

autoMFA: Algorithms for Automatically Fitting MFA Models

Provides methods for fitting the Mixture of Factor Analyzers (MFA) model automatically. The MFA model is a mixture model where each sub-population is assumed to follow the Factor Analysis (FA) model. The FA model is a latent variable model which assumes that observations are normally distributed, but imposes constraints on their covariance matrix. The MFA model contains two hyperparameters; g (the number of components in the mixture) and q (the number of factors in each component Factor Analysis model). Usually, the Expectation-Maximisation algorithm would be used to fit the MFA model, but this requires g and q to be known. This package treats g and q as unknowns and provides several methods which infer these values with as little input from the user as possible. The available methods are a naïve search over both g and q, two different implementations of the AMFA algorithm (Wang and Lin, 2020) <doi = 10.1007/s11749-020-00702-6>, the AMoFA algorithm (Kaya and Salah, 2015) <doi = 10.48550/arXiv.1507.02801> and the VBMFA algorithm (Ghahramani and Beal, 2000) <url = https://mlg.eng.cam.ac.uk/zoubin/papers/nips99.pdf>.

Authors:John Davey [aut, cre], Sharon Lee [ctb], Garique Glonek [ctb], Suren Rathnayake [ctb], Geoff McLachlan [ctb], Albert Ali Salah [ctb], Heysem Kaya [ctb]