Best Practice for Variaional Inference for Gaussian Process

Thanks @fehiepsi That is interesting! Did you use the exact code or you made some changes? How about the MAP (MLE) value of hyper-parameter, are they reasonably close to the MLE results? If everything is fine one your machine it is probably because somehow I got float32.

(Sorry I just don’t have access to my machine now so I can’t do quick runs.)

Thanks a lot!!