Multiple sparse, latent, partially observed Gaussian Processes with variational mini-batch inferencing in the context of a mixed effects model

Misc.
2

I don’t have much experience with GPyTorch but I think it is better suited for your usage case because pyro.contrib.gp does not support learning independent GPs yet. What pyro.contrib.gp supports are:

  • multi-task GP (with task shapes can be arbitrary)
  • minibatch training (see deep kernel learning example or deep gp tutorial)

If independent-GP is not required, then I guess you can use contrib.gp:

  • using obs_mask at sample primitive for partial observation. More flexible solutions can be seen in this topic.
  • the latent variable output can be used for other computation, as in deep gp tutorial