fabMix_missing_values: Function to estimate the UUU or UCU models in case of missing...

In fabMix: Overfitting Bayesian Mixtures of Factor Analyzers with Parsimonious Covariance and Unknown Number of Components

Function to estimate the UUU or UCU models in case of missing values

Description

This function runs parallel chains under a prior tempering scheme of the Dirichlet prior distribution of mixture weights. Missing values are simulated from their full conditional posterior distribution.

Usage

fabMix_missing_values(sameSigma, dirPriorAlphas, rawData, outDir, Kmax, mCycles, 
	burnCycles, g, h, alpha_sigma, beta_sigma, q, normalize,  
	thinning, zStart, nIterPerCycle, gibbs_z, warm_up, 
	progressGraphs, gwar, lowerTriangular)

Arguments

sameSigma	Logical value denoting the parameterization of the error variance per component. If sameSigma = TRUE, the parameterization UCU is fitted, otherwise the UUU model is fitted.
dirPriorAlphas	The prior Dirichlet parameters for each chain.
rawData	The observed data as an n\times p matrix. Clustering is performed on the rows of the matrix.
outDir	Name of the output folder.
Kmax	Number of components in the overfitted mixture. Default: 20.
mCycles	Number of MCMC cycles. Each cycle consists of nIterPerCycle MCMC iterations. At the end of each cycle a swap of the state of two randomly chosen adjacent chains is attempted.
burnCycles	Number of cycles that will be discarded as burn-in period.
g	Prior parameter g. Default value: g = 2.
h	Prior parameter h. Default value: h = 1.
alpha_sigma	Prior parameter \alpha. Default value: \alpha = 2.
beta_sigma	Prior parameter \beta. Default value: \beta = 1.
q	Number of factors q, where 1 \leq q \leq L. An error is thrown if the Ledermann bound (L) is exceeded.
normalize	Should the observed data be normalized? Default value: TRUE.
thinning	Optional integer denoting the thinning of the keeped MCMC cycles.
zStart	Optional starting value for the allocation vector.
nIterPerCycle	Number of iteration per MCMC cycle. Default value: 10.
gibbs_z	Select the gibbs sampling scheme for updating latent allocations of mixture model. Default value: 1.
warm_up	NUmber of iterations that will be used to initialize the models before starting proposing switchings. Default value: 500.
progressGraphs	Logical value indicating whether to plot successive states of the chains while the sampler runs. Default: FALSE.
gwar	Initialization parameter. Default: 0.05.
lowerTriangular	logical value indicating whether a lower triangular parameterization should be imposed on the matrix of factor loadings (if TRUE) or not. Default: TRUE.

Value

List of files written to outDir

Note

It is recommended to always use: normalize = TRUE (default). Tuning of dirPriorAlphas may be necessary to achieve reasonable acceptance rates of chain swaps. Also note that the output is not identifiable due to label switching and the user has to subsequently call the dealWithLabelSwitching function.

Author(s)

Panagiotis Papastamoulis

fabMix documentation built on May 29, 2024, 2:53 a.m.