Practical graph neural networks in Python with TensorFlow and Spektral
summary
Live code shows GNNs implemented with TensorFlow and Spektral, using Colab/GPU, concrete layers GCN and GAT, and dataset plumbing (single vs disjoint loaders). Results include Cora node classification — Cora: 2,708 nodes 1,433 features 7 classes — with a trained model hitting ~0.998 accuracy on the provided split. Also shows graph-level training on PROTEINS with global pooling, a reusable custom Dataset class pattern, and production-ready code used for traffic prediction at TensorCell.
AI-written summary · about
chapters (19)
- 0:00Agenda
- 1:45About me
- 2:57What can you get
- 4:28Environment
- 6:19Graph neural networks
- 11:29Semisupervised classification with graph convolutional networks
- 13:18Example karate club network
- 16:38Graph representation
- 17:07Adjacency matrix
- 19:23Feature matrix
- 25:10Types of neural networks
- 35:35Types of libraries
- 38:01Main components of spectral
- 40:33Practice time
- 43:35Questions
- 47:02Data
- 50:46Features
- 53:01Exercise
- 54:35Loaders
related talks
31mGraph-Based Machine Learning with Python
59mTransforming DataFrames: Intro to Graphs & GNNs with PyTorch Geometric
47mPyTorch Geometric for Graph Neural Nets
18mTC-GNN: Bridging Sparse GNN Computation and Dense Tensor Cores on GPUs
29mGraph Neural Networks + LLM on
19mNeuGraph: Parallel Deep Neural Network Computation on Large Graphs
24mJet Energy Corrections with GNN Regression using Kubeflow @ CERN
32mCutting Edge Football Analytics in Python