Transforming DataFrames: Intro to Graphs & GNNs with PyTorch Geometric
summary
Tutorial demonstrates converting tabular data into graphs with PyTorch Geometric and NetworkX and trains simple GNNs on concrete datasets. It loads a Spotify dump (~300,000 tracks, sampled to 3,000 sampled tracks), builds playlist/genre edges (initially ~1,000,000 reduced to ~800,000 edges), creates node feature tensors and edge index, fits a 2‑layer GCN for track popularity (500 epochs, loss falls but predictions are poor due to heavy downsampling). A hetero example builds a bipartite book–user graph, uses HeteroData, an encoder + edge decoder and shows working end‑to‑end code.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:47What graphs are and why
- 6:07Graph Neural Networks explained
- 11:51Tabular data to graphs
- 16:09Spotify homogeneous graphs
- 34:18Book recommendation graphs
- 53:23Resources and Q&A
related talks
47mPyTorch Geometric for Graph Neural Nets
31mGraph-Based Machine Learning with Python
16mPreview: Graph NNs
29mGraph Neural Networks + LLM on
01h 30mPractical graph neural networks in Python with TensorFlow and Spektral
39mUsing Graph Neural Networks to Embrace the Dependency Within Your Data
01h 01mToward Universal Graph Representations: Foundations and Frontiers of Graph Foundation Models
19mNeuGraph: Parallel Deep Neural Network Computation on Large Graphs