tensorflow/gnn

What it solves

It provides a scalable framework for building Graph Neural Networks (GNNs) on the TensorFlow platform, specifically handling both homogeneous and heterogeneous graphs (graphs with multiple types of nodes and edges).

How it works

The library uses a specialized GraphTensor type to represent complex graph schemas. It includes a graph sampler to convert massive databases into manageable subgraphs for training and inference, a set of pre-built Keras layers and models for GNN architecture design, and a high-level API to orchestrate the training process.

Who it’s for

Machine learning engineers and researchers who need to build and scale GNNs using the TensorFlow ecosystem.

Highlights

• Support for heterogeneous graph schemas with multiple node and edge types.
• Integrated graph sampling tools for processing huge datasets.
• Ready-to-use models and Keras layers for rapid GNN modeling.
• High-level training orchestration API.