Graph neural networks in TensorFlow-Keras with RaggedTensor representation (kgcnn)

Graph neural networks are a versatile machine learning architecture that received a lot of attention recently due to its wide range of applications. In this technical report, we present an implementation of graph convolution and graph pooling layers for TensorFlow-Keras models, which allows a seamless and flexible integration into standard Keras layers to set up graph models in a functional way. We developed the Keras Graph Convolutional Neural Network Python package kgcnn based on TensorFlow-Keras which focus on a transparent tensor structure passed between layers and an ease-of-use mindset. •Python package for graph convolutional neural networks in Tensorflow-Keras using RaggedTensors.•Flexible integration and modular layers for setting up custom graph learning models.•Transparent tensor representation allows readable coding style and makes code easy to debug.•Popular graph neural networks are implemented as an example.•Implemented models expose hyperparameters for easy customization, tuning and experimenting.