Notes on "Regression and Other Stories"

I organized a reading group for "Regression and Other Stories" a statistics textbook by Andrew Gelman, Jennifer Hill, and Aki Vehtari. This repo contains my notes on the reading, chapter by chapter, as well as any demo scripts I wrote to explore the material.

The book is built around the programming language R, but I will be using Python throughout, esp. Jupyter notebooks.

Chapter Notes

• Part 1: Fundamentals
  • Chapter 1: Overview
  • Chapter 2: Data and measurement
  • Chapter 3: Some basic methods in mathematics and probability
  • Chapter 4: Statistical inference
  • Chapter 5: Simulation
• Part 2: Linear Regression
  • Chapter 6: Background on regression modeling
  • Chapter 7: Linear regression with a single predictor
  • Chapter 8: Fitting regression models
  • Chapter 9: Prediction and Bayesian inference
  • Chapter 10: Linear regression with multiple predictors
  • Chapter 11: Assumptions, diagnostics, and model evaluation
  • Chapter 12: Transformations and regression
• Part 3: Generalized linear models
  • Chapter 13: Logistic regression
  • Chapter 14: Working with logistic regression
  • Chapter 15: Other generalized linear models
• Part 4: Before anad after fitting a regression
  • Chapter 16: Design and sample size decisions

Jupyter and Bambi

The exercises I completed were done using Bambi as the Bayesian inference engine, inside Jupyter notebooks.

Originally I could skate by with a virtualenv and pip installing Bambi, but, when I upgraded to Python 3.14 I got bit by some C++ compiler bugs. I had to bite the bullet and use Conda instead.

The steps I took were guided by the above Conda link, and looked like:

Download the Miniconda mini-installer for my Linux (well, WSL) OS:

$ curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

Run that downloaded installer:

$ bash ~/Miniconda3-latest-Linux-x86_64.sh

Which responded with (among lots of other text):

Miniconda3 will now be installed into this location: /home/bgawalt/miniconda3

From there, I set up the Conda environment for this project:

$ conda create --name ros_conda scipy numpy jupyter matplotlib
$ conda install -c conda-forge bambi

That last line was from the Bambi "Getting Started" guide. (I tried just stuffing bambi into that list of args alongside scipy and numpy but that didn't work. But the second line did, so, cool.)

Back when I was on Python 3.10 and using flat virtualenvs and pip to install the Bambi module, I needed to install g++ and python3-dev in order for the NUTS chain sampler to run at tolerable speed:

$ sudo apt-get update
$ sudo apt install python3-dev
$ sudo apt install g++

This made chain sampling go 100x faster. These are still around, but, I suspect they're now ignored in favor of what's been compiled inside the Conda environment.

When it's time to write more exercises, I launch the Jupyter backend (always from the root dir of this repo) with:

$ conda activate ros_conda; \
cd ~/ros; \
jupyter notebook --port 9999

I use that manually-specific port to avoid colliding with Pelican (my static-site blogging software) on :8888.

More details: https://docs.jupyter.org/en/latest/running.html