Probablistic Programming Data Science with PyMC3
summary
Builds and runs Bayesian models in PyMC3 using a Student t likelihood and Hamiltonian Monte Carlo NUTS HMC to draw 5,000 samples, producing a posterior probability of about 3.34% that in‑sample mean exceeds out‑of‑sample mean so the strategy likely overfits. Also implements a Bayesian neural net with ~95.4% holdout accuracy, compares MCMC to ADVI variational inference and mini‑batch ADVI for speed, and shows posterior predictive checks and uncertainty maps.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:10Why probabilistic programming
- 3:15Overfitting in backtesting
- 8:35Frequentist vs white-box modeling
- 11:04Building model with PyMC3
- 19:35Ignoring overfitting consequences
- 23:21Bayesian neural networks example
- 30:14Variational inference
- 33:04Advantages of probabilistic programming
- 34:29PyMC3 learning resources
- 35:57Q&A
related talks
35mVariational Inference and Python
44mAn introduction to PyMC3
42mBayesian Non-parametric Models for Data Science using PyMC3
01h 22mWriting effective bayesian programs using TensorFlow and TFP
01h 00mA Hitchhiker's Guide to designing a Bayesian library in Python
36mClusterf*ck: A Practical Guide to Bayesian Hierarchical Modeling in PyMC3
30mFighting Gerrymandering with PyMC3
14mProbabilistic Programming in Python