This is a re-implementation of the Graph Auto-encoder and Variational Graph Auto-encoder presented here.
This implementation is based on Spektral, the Tensorflow-Keras library for Graph Neural Networks. Compared to the original implementation, the code is more compact and the scripts are self-contained.
The results obtained are slightly different from those reported in the original paper (the AUC is actually higher here). Also, the results show that there is not a significant improvement in using the variational variant compared to the deterministic one.