EandrewJones/mmBPFA: Multi-Modal Beta-Process Factor Analysis

A semi-parametric, multidimensional Bayesian factor analysis for multimodal data. By combining a beta-process prior (Knowles and Ghahramani 2011) on the factor loadings matrix and treating the observed margins as arbitrary manifestations of a latent Gaussian copula (Murray et al. 2013), this package allows the user to perform a factor analysis on data with any combination of marginal distributions (continuous, discrete, or both) without first specifying the dimensionality (K) of the data. The non-parametric prior automatically learns the true dimensionality. Potential applications include traditionaldimension reduction, multidimensional scaling (item-response theory), and missing data imputation. The sampler is implemented in C++ for improved speed and includes post-processing functions that are amenable to a tidy workflow.

install.packages("remotes")
remotes::install_github("EandrewJones/mmBPFA")