Podcast
Bayesian Modeling: PyMC, Stan and Probabilistic Programming Workflows
Open original DataTalks.Club episode
Bayesian Modeling: PyMC, Stan and Probabilistic Programming Workflows
Original Episode
Use these links for the canonical episode and media sources.
- Open the original DataTalks.Club podcast page
- Watch on YouTube
- Listen on Spotify
- Listen on Apple Podcasts
Episode Overview
How do you move from point estimates to full uncertainty-aware models and choose the right tools and workflows for Bayesian modeling? In this episode Rob Zinkov, a machine learning engineer and former Indiana University research scientist who led development of the Hakaru probabilistic programming language, walks through practical Bayesian workflows and tool choices. We cover the core challenge of encoding priors, likelihoods, and posteriors; why integrals become intractable and how numerical integration and.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 0:00 - Episode Introduction & Topic Overview
- 1:44 - Guest Introduction: Rob Zinkov and the Hakaru probabilistic programming project
- 2:46 - Career Journey: From software engineering to machine learning research
- 3:57 - Industry vs Academia: Applying Bayesian tools in real problems
- 6:40 - Transitioning Skills: Embracing calculus, integrals, and optimization
- 8:12 - Core Technical Skills: Linear algebra and optimization for ML
- 9:32 - Self-Study Path: Learning statistics without formal classes
- 14:47 - Statistical Paradigms: Frequentist point estimates vs Bayesian distributions
- 19:06 - Bayesian Workflow: Priors, likelihoods, and posterior distributions
- 21:31 - Bayesian Advantages: Composability and incremental model building
- 23:45 - Probabilistic Programming: Automating Bayesian model tasks
- 24:29 - Why Integrals Matter: Intractable integrals in probabilistic models
- 26:40 - Numerical Integration: Sampling as an approximation technique
- 29:17 - Samplers Overview: Using draws to estimate posterior expectations
- 33:48 - MCMC Fundamentals: Markov chains and exploring high-probability regions
- 36:39 - Probabilistic Languages: Hakaru’s role in generating samplers
- 39:38 - Language vs Library: Model semantics, control flow, and ASTs
- 43:20 - PyMC Example: Building a rainfall model and computational graph
- 48:10 - Interpreting Posteriors: Model checks and iterative refinement
- 51:17 - Encoding Dependencies: Spatial models and hierarchical structure
- 53:12 - Multimodality & Uncertainty: Representing multiple plausible outcomes
- 55:41 - Stan & HMC/NUTS: Advances in efficient sampling algorithms
- 1:00:47 - Learning Resources: PyMC book, Statistical Rethinking course, and tutorials
- 1:05:53 - Consulting & Contact: Rob’s statistical consulting and email