Master Advanced Stan Programming
Go beyond foundations into production-ready Bayesian modeling: mixture models, survival analysis, Gaussian Processes, physics-constrained likelihoods, and performance engineering - all in Stan with a Columbia professor.
Official PartnerEvery student gets access to enterprise cloud virtual machines with Stan pre-configured, directly through the AthlyticZ Academy LMS.
Build on Strong Foundations
This advanced course assumes mastery of foundational Bayesian concepts and R/Tidyverse fluency.
Priors, likelihoods, posteriors, GLMs, hierarchical models, cmdstanr basics.
View Course →Data wrangling, visualization, and functional programming in R.
View Course →The Gap Between Mid-Level and Principal
Overdispersion breaks simple models
Poisson assumes mean equals variance. Real count data doesn’t. Excess zeros, heavy tails, and overdispersion need mixture approaches.
NB, ZIP, ZINB & hurdle models
Principled mixture models that handle overdispersion and excess zeros. The right tool for every count distribution.
Smooth functions need hand-tuning
Polynomial fits oscillate. Arbitrary smoothing parameters have no principled basis. Small changes wreck predictions.
GPs, HSGP & splines in Stan
Gaussian Processes with full posterior uncertainty. Scalable HSGP approximations. B-splines and tensor products for flexible xG models.
Stan models too slow for production
Sampling takes hours. No parallelization strategy. Models that work on toy data fail on real-scale datasets.
Performance engineering for Stan
Vectorization, within-chain parallelization, GPU acceleration, QR reparameterization, and memory management. Make Stan fly.
Scott Spencer
Scott Spencer is one of the world’s foremost experts in applied Bayesian analysis. As a Columbia professor and Stan collaborator, his methods power decisions at Amazon, Johnson & Johnson, Vevo, and leading sports franchises. In this advanced course, Scott shares the production-ready patterns and performance optimizations that distinguish academic exercises from real-world Bayesian systems.
19 Advanced Modules
19 modules. 136 lessons. 30+ Stan files. Click any module to see every lesson.
Establish a repeatable Bayesian workflow, then tackle overdispersion and excess zeros with NB, ZIP, ZINB, and hurdle models using baseball and basketball data.
Rigorous Workflow Before Fancy Models
Advanced models are worthless without rigorous workflow. Scott starts with the end-to-end process - prior predictive checks, diagnostics, posterior validation, model comparison - before touching any advanced technique. This discipline is what separates production Bayesian from academic Bayesian.
Rating and ranking models, advanced hierarchical structures, sufficient statistics, correlation modeling, QR decomposition, and autoregressive processes.
Computational Tricks That Make Stan Practical
QR decomposition fights collinearity. Sufficient statistics compress data without losing information. Copulas model flexible dependence structures. These are the computational tricks that turn slow, unstable Stan models into fast, reliable production systems.
The core of the advanced curriculum: survival analysis, differential and difference equations, B-splines, full Gaussian Processes, and scalable HSGP approximations.
Survival. GPs. Splines. The Skills That Command $250K+
These six modules cover the techniques that separate a $120K data scientist from a $250K+ principal. Pfizer uses survival models for drug approval. Citadel uses GPs for risk pricing. Amazon uses HSGP for personalization at billion-user scale. Every technique here maps to a real hiring signal.
Embed mechanics into likelihoods for sailing, golf, and baseball. Handle missing data, censoring, and outliers. Then make Stan fly with vectorization, parallelization, and GPU acceleration.
Physics in the Likelihood. Production at Scale.
By the end of this course, you can embed sailing mechanics into a Stan model, impute missing tracking data, handle censored observations, and make your models fast enough for production with GPU acceleration and within-chain parallelization. This is the full advanced Bayesian toolkit.
Trusted by Analytics Leaders
“Scott Spencer brings Bayesian modeling alive. He not only explains the math, but shows how to implement models in Stan that are clear, scalable, and ready for research or production.”
“AthlyticZ has completely transformed the learning approach to data science through the use of sports-based problems. The instructors are the best of the best.”
Enroll in Becoming a BayeZian II
Becoming a BayeZian II
- 19 modules, 136 lessons
- 30+ advanced Stan model files
- GPs, HSGP, survival, splines, physics models
- Performance engineering: GPU, parallelization
- Taught by Columbia professor Scott Spencer
- Certificate of completion
Stan pre-configured via AthlyticZ Academy LMS
Members pay $1,159 • Explore Membership
Get Both Bayesian Courses Together
Combine Becoming a BayeZian I (foundations) + Becoming a BayeZian II (advanced) into one bundle for a complete Bayesian education - from priors to physics-constrained models.
Members Pay $1,159 for This Course
Plus unlimited cloud VM access, 10+ live sessions per month, the full replay library, and 20% off every course in the catalog.
Frequently Asked Questions
Join the Elite Analysts Who
Build What Others Can’t
19 modules. 136 lessons. 30+ Stan files. GPs, survival, physics-constrained models. Cloud virtual machines included.