Keywords: library, edge representation learning, pytorch
Abstract: Machine learning on graphs (GraphML) has been successfully deployed in a wide variety of problem areas, as many real-world datasets are inherently relational. However, both research and industrial applications require a solid, robust, and well-designed code base. In recent years, frameworks and libraries, such as PyTorch-Geometric (PyG) or Deep Graph Library (DGL), have been developed and become first-choice solutions for implementing and evaluating GraphML models. These frameworks are designed so that one can solve any graph-related task, including node- and graph-centric approaches (e.g., node classification, graph regression). However, there are no edge-centric models implemented, and edge-based tasks are often limited to link prediction. In this extended abstract, we introduce PyTorch-Geometric Edge (PyGE), a deep learning library that focuses on models for learning vector representations of edges. As the name suggests, it is built upon the PyG library and implements edge-oriented ML models, including simple baselines and graph neural networks, as well as corresponding datasets, data transformations, and evaluation mechanisms. The main goal of the presented library is to make edge representation learning more accessible for both researchers and industrial applications, simultaneously accelerating the development of the aforementioned methods, datasets and benchmarks.
Type Of Submission: Extended abstract (max 4 main pages).
Agreement: Check this if you are okay with being contacted to participate in an anonymous survey.
PDF File: pdf
Type Of Submission: Extended abstract.
Software: https://github.com/pbielak/pytorch_geometric_edge/
Poster: png
Poster Preview: png
6 Replies
Loading