Benefits of Being Bayesian: Motor Imagery EEG Classification (Thesis Software)

Software accompanying the MS thesis Benefits of Being Bayesian: Motor Imagery Electroencephalogram Classification (University of Washington, 2026), which asks whether Bayesian machine learning measurably improves left/right hand motor-imagery electroencephalogram (MI-EEG) classification over matched frequentist baselines. The repository evaluates twenty MOABB datasets under cross-subject cross-validation across six paired frequentist/Bayesian pipelines, spanning common spatial pattern, tangent space, and convolutional architectures, with posteriors sampled by the No-U-Turn Sampler in PyMC. Six metrics cover discrimination (AUROC, MCC), calibration (ECE, MCE), and proper scoring (NLL, Brier score), with per-pipeline compute profiling via CodeCarbon. Effect sizes are pooled in three-level meta-analyses with robust variance estimation and meta-regression on sample size.