Examples & Case Studies¶
Sixteen executable notebooks demonstrating Aurora-GLM across GLM, GAM, and GAMM workflows. Each case study walks through data preparation, model specification, fitting, diagnostics, and interpretation. Every model mentioned in a notebook is actually fitted in it. All notebooks are available on GitHub and can be run locally.
Generalized Linear Models (GLM)¶
Medical insurance charges with a Gamma GLM: log vs identity link comparison, age × BMI interaction, and risk segmentation.
Claim frequency with an exposure offset: overdispersion detection and Negative Binomial with ML-estimated dispersion.
Logistic regression with odds ratios, ROC analysis, and calibration diagnostics.
Beta regression for bounded scores, with spline regression for the non-linear age effect.
Multi-model Poisson regression with rate ratios and nested-model comparison.
Large-n binary classification with calibration and class-imbalance analysis.
Generalized Additive Models (GAM)¶
Additive model with GCV smoothing-parameter selection and identifiability constraints.
Hourly rental counts with smooths for temperature, humidity, and time-of-day effects.
Gamma GLM baseline vs GAM on log(power), with a power-curve analysis.
Conversion rates with smooth price, duration, and hour effects (PQL) plus a random intercept per day-of-week.
Generalized Additive Mixed Models (GAMM)¶
Random intercepts and slopes on the classic sleep-study dataset, with marginal vs conditional R².
Repeated measures with an identity vs AR(1) covariance comparison and effect sizes.
Patients nested in clinics; the LPM-with-random-effects vs PQL trade-off under separation.
Crossed random effects for subjects and items — the large sparse mixed-model case.
Eight models from null to random slopes on real PISA UK data, with variance partitioning.
Multi-center binary outcome via Penalized Quasi-Likelihood: odds ratios, latent-scale ICC, and PQL limitations.