Benchmarking GNNs with GenCAT Workbench

Seiji Maekawa, Yuya Sasaki, George Fletcher, Makoto Onizuka

Abstract: We present GenCAT Workbench, an end-to-end framework with which users can generate synthetic attributed graphs with node labels and evaluate their graph analytic methods, e.g., graph neural networks (GNNs), on the generated graphs. GenCAT Workbench supports various types of graphs with controlled node attributes and graph topology. We demonstrate the GenCAT Workbench and how it clarifies the strong and weak points of GNN models. Our code base is available on Github ( https://github.com/seijimaekawa/GenCAT/tree/main/GenCAT_Workbench ).