Frameworks for deep learning on graphs
Over the years, a few libraries have been developed to help data scientists integrate graph machine learning, and more specifically, GNNs, into their analytical pipelines. The existence of different libraries is due to different choices of the framework used in the backend (e.g., TensorFlow, PyTorch, etc.), the research group authoring the library (sometimes being an industrial group and sometimes an academic department), as well as some specific needs addressed by the library (e.g., large graphs, dynamic graphs, etc.).
In the following, we will consider three frameworks:
- PyTorch Geometric (PyG):
As the name suggests, PyG (https://pytorch-geometric.readthedocs.io/en/latest/) is a library that is built upon PyTorch, and it provides abstractions and functionalities for defining, training, and evaluating GNNs. More generally, it provides various methods for performing deep learning on graphs and irregular structures. It also...
