Lies damned lies and statistics in Python
summary
Compares logistic regression implementations in Statsmodels scikit-learn and PyMC3 on the UCI Adult dataset, turning income into a >50k binary target (24% positive) and reporting ~79–80% accuracy but a pseudo R² ≈ 0.009; shows machine‑learning coefficients, Statsmodels summaries, and PyMC3 Bayesian posteriors using Metropolis and NUTS, notes slow sampling (minutes), convergence problems on some parameters, explains priors, provides notebooks and a GitHub repo for reproducing the experiments.
AI-written summary · about
chapters (9)
- 0:00Statistics pitfalls in Python
- 3:46Logistic regression approaches
- 8:04Logistic regression demo
- 18:31Bayesian A/B testing with PyMC3
- 23:11Frequentist hypothesis testing
- 35:24Real data model applications
- 44:53Debugging PyMC3 models
- 53:08Optimizing PyMC3 performance
- 1:01:09Q&A
related talks
31mLogistic Regression Behind the Scenes
40mMachine Learning Model Evaluation Metrics
57mA Quickstart Guide to PyMC3
42mBayesian Non-parametric Models for Data Science using PyMC3
01h 26mProbabilistic Programming and Bayesian Computing with PyMC
43mProbabilistic Programming in the Real World
14mProbabilistic Programming in Python
36mClusterf*ck: A Practical Guide to Bayesian Hierarchical Modeling in PyMC3