spektral

Why we included this project

Graphs describe a lot of real-world data that tabular and sequence models handle awkwardly: social connections, molecular structures, and the links between documents or users. Spektral is a Python library built on the Keras API and TensorFlow 2 that brings graph neural networks to that stack, so teams already comfortable with Keras can train GNNs without adopting a separate framework. It bundles a broad set of established layers, including GCN, GraphSAGE, GAT, and GIN, plus pooling layers and utilities for transforming graph data into the tensors Keras expects. That covers common jobs like node classification, link prediction, and predicting molecular properties, all inside the standard Keras model-and-fit flow. The documentation and example templates make it easy to get a first model running and to see how graph inputs should be structured before you design your own.