// the find
mlabonne/graph-neural-network-course
Free hands-on course about Graph Neural Networks using PyTorch Geometric.
A four-notebook crash course on Graph Neural Networks built on PyTorch Geometric, covering GCN, GAT, GraphSAGE, and GIN. Aimed at people who already know standard deep learning and want a hands-on, code-first intro to graphs rather than a textbook treatment.
Each notebook pairs a runnable Colab link with a companion blog post that explains the math, so you get both the 'why' and the 'how' instead of just copy-pasteable code. The progression is deliberate: plain GCN on CiteSeer, then attention, then mini-batching for scale with GraphSAGE on PubMed, then graph-level classification with GIN on PROTEINS — each notebook introduces exactly one new idea. Written by Maxime Labonne, who has a track record of clear ML explainers, so the writing quality is higher than the average tutorial repo.
Dead since August 2023 — no updates for newer architectures (no GraphTransformer, no heterogeneous graph support, nothing on scaling tricks introduced since). Only 4 notebooks total, so 'course' overstates the scope; it's a solid intro and nothing past that. There's no requirements.txt or environment pin, which matters for notebooks — PyTorch Geometric's API has had breaking changes since 2023, so these may not run as-is without dependency surgery. The repo itself is just notebooks plus a README; the real content (explanations) lives off-repo on the author's blog, so GitHub alone isn't self-contained if those links ever rot.