Welcome!

In this workshop, you'll learn how to develop Bayesian intuition and build powerful probabilistic models using PyMC.

Making decisions under uncertainty is hard — especially when your data is limited, your outcomes are rare, or your assumptions are hidden. 😭

You'll see how modern Bayesian modeling can:

• Estimate probabilities with informative priors.
• Compare alternatives probabilistically with Bayesian A/B testing.
• Share strength across groups using hierarchical models.
• Evaluate and anticipate rare events using posterior predictive distributions.

You can run the notebooks locally using PyMC, ArviZ, and Jupyter Notebooks — or on Colab with no setup required.

This workshop is based on tutorials taught my PyMC Labs, with some examples from Allen Downey's book Think Bayes.

---

Use these links to run the notebooks on Colab (no setup required):

🚀 Notebook 1: Priors and Posteriors

🎯 Notebook 2: Bayesian Bandits

🧬 Notebook 3: Hierarchical Models

🔮 Notebook 4: Probabilistic Prediction

Or follow the instructions below to run the notebooks locally.

Outline

✅ Phase 1: Building Bayesian Intuition

You'll start by developing a solid foundation in Bayesian thinking:

• Understand the fundamentals of probability and uncertainty.
• Learn how to specify informative priors based on domain knowledge.
• Build simple models to estimate unknown rates from sparse observations.
• Practice interpreting posterior distributions and credible intervals.

This gives you the Bayesian mindset needed for more complex modeling.

---

✅ Phase 2: Probabilistic Decision Making

After building Bayesian intuition, you'll create decision-support systems:

• Use Bayesian A/B testing to compare alternatives probabilistically.
• Build hierarchical models to pool data across many subgroups.
• Identify and avoid catastrophic sequences of failures.
• Evaluate whether your models make robust predictions under uncertainty.

You'll also learn how to:

• Balance generalization and specificity in your models.
• Communicate risks and confidence clearly to stakeholders.

---

✅ Phase 3: Advanced Bayesian Workflows

Finally, you'll move from individual models to complete Bayesian workflows:

• Build posterior predictive distributions to evaluate rare events.
• Develop systematic approaches to model validation and criticism.
• Integrate Bayesian thinking into your broader modeling workflow.
• Reason clearly, act decisively, and manage uncertainty with confidence.

---

Running Locally

We recommend you run the notebooks on Colab, using the links above. But if you want to run locally, you'll need a working Python 3.11+ environment with PyMC and related packages. We've provided requirements files for Conda/pip, pixi, and uv, and we recommend them in that order.

1. Setup with Conda/pip

Navigate to the repository and use the Makefile to create a Conda environment and install the requirements:

cd ai_decision_workshop
make create_environment
conda activate ai_decision_workshop
make requirements

2. Setup with pixi

Navigate to the repository and activate the environment:

cd ai_decision_workshop
pixi shell

3. Setup with uv

Note: This option requires separate installation of GraphViz, which is used to display graphical representations of the models.

Navigate to the repository and create the environment:

cd ai_decision_workshop
uv venv
source .venv/bin/activate  # macOS/Linux
# or
.venv\Scripts\activate  # Windows
uv pip install -r requirements.txt

Install GraphViz:

• macOS: brew install graphviz
• Ubuntu/Debian: sudo apt-get install graphviz
• Windows: Download from GraphViz website

---

• Notebooks are located in the notebooks/ folder.
• Solutions are in the soln/ folder.

Start with the notebooks in the notebooks/ folder and work through them in order.

---

Ready to build your Bayesian decision-making skills? Let's go. 🚀