Hierarchical Probabilistic Modelling in Real Life
summary
Claims a working pipeline that uses PyMC3 and NUTS sampler to turn realized spend into budget ranges, testing three models (pooled, product-level, product+region) on a derived real dataset with five products. WAIC and posterior predictive checks show SEM and SEO match well while display is bimodal and poorly fit; region hyperpriors cause sampler instability. The end result is a Flask app with precomputed predictions (models saved to S3 via Theano shared variables) that returns actionable spend ranges for sales.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 15:01Probabilistic modeling in marketing
- 22:37Bayesian analysis appeal
- 26:07Hierarchical models with examples
- 34:47Marketing data overview
- 39:31Running models in code
- 48:14Markov chain Monte Carlo
- 56:52Pooled model results
- 1:01:22Hierarchical outputs
- 1:06:28Model evaluation metrics
- 1:09:57Posterior predictions
- 1:14:36Model fit checks
- 1:17:54Prediction testing
- 1:20:29Production pipeline
- 1:24:26Q&A
related talks
01h 26mProbabilistic Programming and Bayesian Computing with PyMC
42mSolving Real-World Business Problems with Bayesian Modeling
01h 27mProbabilistic Python: An Introduction to Bayesian Modeling with PyMC
35mWhat we learned by running a large custom Bayesian...
43mProbabilistic Programming in the Real World
39mProbablistic Programming Data Science with PyMC3
28mInterpretable and realistic generative models in data science?
39mHow to Allocate Marketing Budget - Bayesian Approach to the Rescue