https://github.com/tallamjr/bap-numpyro
Bayesian Analysis in Python (2nd ed.) with Numpyro
https://github.com/tallamjr/bap-numpyro
bayesian-data-analysis bayesian-inference laplace-approximation markov-chain-monte-carlo numpyro variational-inference
Last synced: 20 days ago
JSON representation
Bayesian Analysis in Python (2nd ed.) with Numpyro
- Host: GitHub
- URL: https://github.com/tallamjr/bap-numpyro
- Owner: tallamjr
- License: mit
- Created: 2021-09-15T11:37:47.000Z (almost 5 years ago)
- Default Branch: master
- Last Pushed: 2026-06-23T00:32:25.000Z (20 days ago)
- Last Synced: 2026-06-23T02:15:21.669Z (20 days ago)
- Topics: bayesian-data-analysis, bayesian-inference, laplace-approximation, markov-chain-monte-carlo, numpyro, variational-inference
- Language: Jupyter Notebook
- Homepage: https://tallamjr.github.io/bap-numpyro/
- Size: 63.6 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Bayesian Analysis in Python (2nd ed.) with Numpyro
After discovering the fantastic
[`rethinking-numpyro`](https://fehiepsi.github.io/rethinking-numpyro/) project by
[@fehiepsi](https://github.com/fehiepsi), I was inspired to try and do something like that myself.
Primarily as a learning activity, this is my attempt at porting
[@aloctavodia](https://github.com/aloctavodia)'s "Bayesian Analysis in Python" [example PyMC3
code](https://github.com/aloctavodia/BAP) to [NumPyro](https://github.com/pyro-ppl/numpyro).
The port is now complete and modernised for 2026: all chapters (1-8) and exercises are ported, including newly-authored exercises for chapters 5-8 that do not exist in the original BAP repository.
## Setup
The project is managed with [uv](https://docs.astral.sh/uv/) and targets the following stack:
| Tool | Version |
|---------|---------|
| Python | 3.13 |
| NumPyro | 0.21 |
| JAX | 0.10 |
| ArviZ | 0.23 |
Create the environment:
```bash
uv sync --group dev
```
Run all notebooks as tests:
```bash
uv run pytest --nbmake notebooks/
```
Build the book locally:
```bash
uv run jupyter-book build .
```
### Running interactively in Jupyter Lab
The notebooks are configured to use a dedicated `bap-numpyro` kernel that points at the project's `.venv` interpreter. Register it once:
```bash
uv run python -m ipykernel install --user --name bap-numpyro --display-name "Python (bap-numpyro)"
```
Then launch Jupyter from the project environment and open any notebook:
```bash
uv run jupyter lab
```
Each notebook already selects the "Python (bap-numpyro)" kernel. Do not run them with a generic "Python 3" kernel: that may resolve to a different environment that lacks the pinned NumPyro/ArviZ stack and will raise import errors. If you see `ModuleNotFoundError: No module named 'numpyro'` (or `arviz`), the wrong kernel is selected: switch to "Python (bap-numpyro)" via Kernel > Change Kernel.
**Graphviz:** the model-rendering cell in the Chapter 2 exercises requires the `graphviz` system package. Install it with `brew install graphviz` (macOS) or `apt-get install graphviz` (Debian/Ubuntu).
**Float64:** the Gaussian Process chapter and its exercises call `numpyro.enable_x64()` for numerical stability. Ensure JAX float64 is not disabled in your environment.
## Acknowledgements
Many many thanks to [Du Phan](https://github.com/fehiepsi) for the foundations of which I have built
this from and [Osvaldo Martin](https://github.com/aloctavodia) for the amazing book: Bayesian Data
Analysis in Python