A GNN workbook that admits its own typos and version conflicts

What it does

This repository holds the chapter-by-chapter Jupyter notebooks for Hands-On Graph Neural Networks Using Python, a Packt book by J.P. Morgan senior applied researcher Maxime Labonne. The code walks through graph neural network fundamentals up through applied tasks—node, graph, and edge classification, topology prediction, and temporal event forecasting—using PyTorch Geometric. It is strictly a companion workbook; you will need the book for explanations.

The interesting bit

The repo does not pretend to be a polished framework. Instead, it offers a curated, chapter-ordered path from basic graph theory to heterogeneous and temporal GNNs, backed by an explicit errata page that corrects formulas and even indegree/outdegree definitions. That honesty is refreshing for a publisher repository.

Key highlights

• Covers node, graph, and edge classification, plus graph topology generation and temporal forecasting.
• Notebooks are organized by chapter and can be imported directly into Google Colab.
• Includes a downloadable PDF of color diagrams and screenshots from the book.
• Maintains a public errata list correcting formulas and terminology errors.
• Authored by a practitioner with a Ph.D. in machine learning and experience in finance and network security.

Caveats

• The dependency matrix is fragile: Chapter 11 requires TensorFlow 2.4, Chapter 14 needs an older PyTorch Geometric (2.0.4), and Chapters 15–17 are memory-intensive enough that the README suggests shrinking the training set. You will likely need multiple virtual environments.
• This is a commercial book companion, not a standalone library or self-contained course.

Verdict

Grab it if you own the book and want ready-to-run PyTorch Geometric examples across a broad GNN curriculum. If you are hunting for a reusable open-source package or a tutorial that stands alone without the text, look elsewhere.