FABLE (Factor Analysis with the BLEssing of dimensionality)

R package to implement Factor Analysis with BLEssing of dimensionality (FABLE) approach. Provides posterior samples of the covariance matrix modeled using factor analysis, along with providing posterior mean (without carrying out sampling).

All results in the paper can be replicated with scripts in extras/replicationCodes .

Paper: https://arxiv.org/abs/2404.03805# .

Install

Please install the devtools package in R.

Usedevtools::install_github("shounakch/FABLE") to install the package.

Example Usage

library(FABLE)
set.seed(1)
n = 500
p = 1000
lambdasd = 0.5
pi0 = 0.5
k = 10
  
Lambda = matrix(rnorm(p*k, mean = 0, sd = lambdasd), nrow = p, ncol = k)
BinMat = matrix(rbinom(p*k, 1, 1-pi0), nrow = p, ncol = k) 
Lambda = Lambda * BinMat
  
Sigma0 = runif(p, 0.5, 5)
  
M = matrix(rnorm(n*k), nrow = n, ncol = k)
E = matrix(rnorm(n*p), nrow = n, ncol = p)
E = sweep(E, 2, sqrt(Sigma0), "*")
  
Y = (M %*% t(Lambda)) + E
    
FABLEPostMean = FABLEPosteriorMean(Y, gamma0 = 1, delta0sq = 1, maxProp = 0.95)
FABLESamples = FABLEPosteriorSampler(Y, gamma0 = 1, delta0sq = 1, maxProp = 0.95, MC = 1000)