factorverse | Peter Dunson

factorverse fits a variety of linear Gaussian Bayesian factor models under different shrinkage and sparsity priors on the loading matrix. The samplers are Gibbs samplers written in C++ via RcppArmadillo, and every method shares one interface, so switching priors is a one-word change.

The model is the linear factor model,

\[\mathbf{x}_i = \boldsymbol{\Lambda}\mathbf{b}_i + \boldsymbol{\epsilon}_i, \qquad \boldsymbol{\epsilon}_i \sim N(\mathbf{0}, \boldsymbol{\Sigma}),\]

with loadings $\boldsymbol{\Lambda}$ ($P \times K$), factor scores $\mathbf{b}_i$, and diagonal residual variances $\boldsymbol{\Sigma}$. The quantity identified without rotation or sign ambiguity is the induced covariance $\boldsymbol{\Omega} = \boldsymbol{\Lambda}\boldsymbol{\Lambda}^{\top} + \boldsymbol{\Sigma}$.

Usage

library(factorverse)

# Simulate from a block-sparse loading structure
Lambda <- sim_lambda(P = 12, K = 3, val = 1.2)
Sigma  <- diag(rep(0.3, 12))
Omega  <- tcrossprod(Lambda) + Sigma
dat    <- sim_data(list(Lambda = Lambda, Sigma = Sigma, Omega = Omega),
                   N = 300, seed = 42)

# Fit a model
fit <- fit_bfm(method = "dl", dat = dat, k0 = 3,
               chains = 1, nrun = 10000, burn = 5000,
               save_rds = FALSE, verbose = FALSE)

Available priors are "mgps", "ssl", "dl", "hs", "bass", and "mnl". The fitted object holds the posterior draws in $posterior and the posterior-mean loadings in $Lambda_hat.

Simulation and diagnostics

sim_run() drives a full replicated study: it builds the parameter sets, generates datasets, fits every method to each one, and records per-replicate metrics, including wall-clock time and effective sample size. sim_summarize() collapses the output into summary tables. At high dimension the fitter switches to a memory-light streaming path so large-$P$ runs stay feasible.

For real data, analyze_factor_fits() and score_correlation() summarize residual and factor correlation structure, bfm_convergence() reports $\hat{R}$ and bulk/tail effective sample sizes, and residuals_by_sample() gives per-draw residual diagnostics.

Install

devtools::install_github("peterdunson/factorverse")

Source on GitHub. Developed in the Crainiceanu Lab at Johns Hopkins Biostatistics. MIT licensed.