Posted On: Dec 3, 2019
Though GNNs have shown promising results in research, their use in real-world applications has been limited because of the complex infrastructure required to train large graphs and the lack of reliable domain specific models. Developing GNNs involves finding and training on very large graphs with millions of nodes, and it is time-consuming to build and maintain the computational infrastructure required to perform this training. DGL gives you the tools and infrastructure to simplify the implementation and deployment of GNNs.
DGL support in Amazon SageMaker removes the burden of packaging software dependencies, building infrastructure, and finding validated models. As a result, you can test and implement GNNs in hours instead of weeks or months. A Deep Learning container bundles all the software dependencies and the Amazon SageMaker API automatically sets up and scales the infrastructure required to train graphs. With the bundled library of validated models, you can immediately test state-of-the-art GNN models and integrate them into applications.