2

I have trained a model in GPflow and ultimately I would like to take this posterior distribution and use it as the prior in a new instance. I reviewed the docs and couldn't see anything. I did see the following link which sounded very good, but it appears to be a placeholder and is empty.

joel
  • 6,359
  • 2
  • 30
  • 55
TZnings80
  • 53
  • 3

1 Answers1

0

There is a difference between "using the posterior as a new prior" and "adding more data to the model". The former is a bit more of a research question, see for example https://github.com/thangbui/streaming_sparse_gp (GPflow 1.0) and https://github.com/pmorenoz/RecyclableGP (PyTorch+GPy). The latter is easy to do (for GPflow's VGP model, see the implementation in the Trieste Bayesian Optimization library built on top of GPflow).

STJ
  • 1,478
  • 7
  • 25