How to apply decoupled sampling for sparse variational GP

[DavidLeeftink]

Hi all,

I am interested in applying decoupled sampling for Sparse Variational GPs (Wilson 2020), however so far I only found the implementation for exact GPs (https://docs.jaxgaussianprocesses.com/examples/spatial/). It seems the 'sample_approx()' function is only part of the ConjugatePosterior() class, but not of the VariationalGaussian() class. Is there currently an implementation for this available?

Much thanks,
David

[DavidLeeftink]

I found it is already implemented and very simple to apply. Since the VariationalGaussian() class has a ConjugatePosterior(), decoupled sampling can simply be done by calling 'svgp_posterior.posterior.sample_approx()', instead of 'svgp_posterior.sample_approx()'