Tutorials on Graph Neural Networks for the DHS 2019 hack session.
Slides: http://bit.ly/dhs2019gnn
Teaser: Labeling one point from each class (the ones with the grey border), and watching the network learn in real-time:
I have included three notebooks.
-
Graph Neural Networks from Scratch is meant to be a very gentle and lightweight introduction to graph neural networks by taking a toy example of predicting the trajectory of planets in a solar system.
-
GCN Karate Club extends https://github.com/tkipf/gcn to visualize, in real time how the karate club node embeddings change during semi-supervised training, when only one point from each community is labelled.
-
GCN Cora extends https://github.com/tkipf/gcn with snippets to visualize the graph, and t-sne visualizations of node embeddings, before and after the semi supervised training.