stellargraph

Why we included this project

StellarGraph is a Python library for training machine learning models on data that comes as graphs and networks, the kind of structure you see in social connections, citation networks, or knowledge graphs. It bundles ready-to-use implementations of popular graph neural network approaches, including graph convolutional networks, GraphSAGE-style aggregation, and attention-based methods. That lets you do node or edge classification, whole-graph classification, and link prediction without building the neighborhood-sampling machinery yourself. It runs on TensorFlow 2 with Keras, so models drop into familiar training loops and work alongside standard Keras layers and scikit-learn workflows. The library also handles heterogeneous graphs and knowledge graphs with multiple edge types, which matters when your relational data is richer than a single homogeneous network. The narrated demo notebooks, runnable in Colab or Binder, are a practical way to match a use case to the right algorithm before committing to a production pipeline.