GaussianProcessClassifier_sonar notebook is showing an exact bianry gaussian process classification algorithm (Rasmussen and Williams, 2006) using tensorflow and tensorflow-probability. This is applicable when the dataset is relatively small. A detailed theoretical explanation can be found https://pengluyaoyao.github.io/post/gpc/
VGPC notebook shows how the sparse variational gaussian process applied to multi-class classification using tensorflow-probability. VGP is especially useful when the dataset is large in which the exact GPC is intractable. The VGP model can be trained batch by batch.