GPLVM with Student-t likelihood

[IvanDeBoi]

Hello GPy enthousiasts,

Can someone tell me how to implement a GPLVM with a Student-t likelihood?

I was hoping for something as easy as this:
m = GPy.models.GPLVM(points, input_dim, kernel=k_GPLVM, init = init)
t_dist = GPy.likelihoods.StudentT(deg_free=5., sigma2=2.)
inference_method = GPy.inference.latent_function_inference.Laplace(), # Laplace approximation for inference
m.likelihood = t_dist;
m.inference_method = inference_method;
These lines work, but then the optimization can not start. Also, changing to the Bayesian GPLVM doesn't help (although different errors pop up).

Seems possible, but I think somewhere in the GPy code (of which I'm certainly not an expert), a Gaussian likelihood is assumed. At least that is what the source code looks like to me.

Can you build a GPLVM based on a GP's you specified beforehand?

Any thoughts / help on this is highly appreciated.

Kind regards,

Ivan