This is the repository for the paper "Automated Augmented Conjugate Inference for Non-Conjugate Gaussian Process Models"
You need Julia 1.1 to run the experiments as well as GPFlow and Tensorflow.
All the source code is implemented in the package AugmentedGaussianProcesses.jl which this experiment package relies on.
To install all the required packaged, go to the directory and run julia:
julia>] activate .
AutoConjugate> instantiate
This will install all the Julia packages needed for the experiments You can then run one of the experiments from the paper:
For the sampling experiment run include("scripts/sampling_exp.jl")
For the hyperparameter experiment run include("scripts/hyperparam_exp.jl")
And for the convergene the code is described in script/_conv_exp.jl
There is a WIP to create a smart macro to create augmented likelihoods from a formula like p(y|f,β) = 1/2β * exp(-sqrt(y^2-2*y*f+y^2)/β), it should arrive in the next weeks