Sparse regression: Lasso vs. Bayesian methods under correlation
New benchmark compares six sparse regression methods (OLS, Ridge, Lasso, Elastic Net, Horseshoe, Spike-and-Slab) across 2,600+ experiments with correlated features, weak signals, and high dimensionality.
• Penalized estimators (Lasso) fit in milliseconds but lack uncertainty estimates
• Bayesian methods (Horseshoe, Spike-and-Slab) yield full posteriors but require minutes per MCMC fit
• Tests vary feature correlation (ρ up to 0.9), signal-to-noise ratio, and feature count (20–100)