Ability to "Freeze" parameters in a GP model?

[mathDR]

Hello! Is there a way to "freeze" the ability of the optimizer to optimize parameters of the GP? In particular, you could imagine we fit the GP to data then want to use that model as part of a larger model in jax. But when we optimize the larger model, we want to keep the parameters of the GP "fixed".

Is there a performant way to do this in GPJax?

In for example, in BayesNewton the constructor for a kernel takes inputs called fix_variance, fix_lengthscale, etc. that tells the model if those parameters are to be included in gradient calculations.

Thanks!

[thomaspinder]

Hi!

Yes! You can do the below

import gpjax as gpx
from gpjax.parameters import Static

sigma_true = 0.2
likelihood = gpx.likelihoods.Gaussian(num_datapoints=n, obs_stddev=Static(sigma_true))