Not Half Bad: Exploring Half-Precision in Graph Convolutional Neural Networks