GNNkeras: A Keras-based library for Graph Neural Networks and homogeneous and heterogeneous graph processing

Abstract

In several areas of science and engineering, data can be naturally represented in graph form, where nodes denote entities and edges stand for relationships between them. Graph Neural Networks (GNNs) are a well-known class of machine learning models for graph processing. In this paper, we present GNNkeras, a library, based on Keras, which allows the implementation of a large subclass of GNNs. GNNkeras is a flexible tool: the implemented models can be used to classify/cluster nodes, edges, or whole graphs. Moreover, GNNkeras can be applied to both homogeneous and heterogeneous graphs, exploiting both inductive and mixed inductive–transductive learning, and can implement a layered version of GNNs, namely the LGNN model.

Keywords

• Graphs
• Machine Learning
• TensorFlow
• Keras
• Graph Neural Networks