getStuffForDIC: Compute information criteria

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

Compute information criteria

Description

This function computes four information criteria for a given run of the fabMix algorithm, namely: AIC, BIC, DIC and \mbox{DIC}_2. Given various runs with different number of factors, the selected model corresponds to the one with the smalled value of the selected criterion.

Usage

getStuffForDIC(sameSigma, sameLambda, isotropic, x_data, outputFolder, q, burn, 
				Km, normalize, discardLower)

Arguments

sameSigma	Logical value indicating whether the parameterization with the same variance of errors per component is used. Default: TRUE.
sameLambda	Logical value indicating whether the parameterization with same loadings per component is used. Default: FALSE.
isotropic	Logical value indicating whether the parameterization with isotropic error variance per component is used. Default: FALSE.
x_data	Observed data.
outputFolder	Name of the folder where the fabMix function has saved its output.
q	Number of factors. Note that this should coincide with the number of factors in the fabMix run.
burn	Discard observations as burn-in period (optional).
Km	Number of components in the overfitted mixture model. Note that this should coincide with the same entry in the fabMix run.
normalize	Should the observed data be normalized? Note that this should coincide with the same entry in the fabMix run. Default value: TRUE.
discardLower	Discard draws with log-likelihood values lower than the specific quantile. This applied only for the DIC computation.

Value

The information criteria are saved to the informationCriteria_map_model.txt file in the outputFolder.

Note

It is well known that DIC tends to overfit, so it advised to compare models with different number of factors using AIC or BIC. The main function of the package uses BIC.

Author(s)

Panagiotis Papastamoulis

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