Okay i have implemented the interpolation function like in the example and that works. I am trying to use the GP priors to do the interpolation
comp=interpolate(Xf,f,X)
likelihood = pm.Normal('y', mu=comp, sigma=yerr, observed=y)
The main reason i am looking for a way to generate points using a conditional distribution is due to efficiency. Defining a bunch of points for the GP prior is expensive computationally, so if i could create a lot of points from the current guess of the hyper-parameters, i believe it would speed up my sampling. This is also the simplest working example i can think of, for my actual use case i am using the GP as an input for a function, which i want to be continuous or at least defined at a lot of points over the range i am considering, so that the interpolation i do with the output of the function using the GP is accurate.