VSGP with separate kernels

I’m trying to make a Gaussian Process that has separate Kernels, meaning a separate lenghtscale and variances parameters per GP. Similar to how in GPLVM one trains D number of GPs for a D dimensional dataset, but in this case, each dimension might have different lengthscales and variances. I see the current implementation makes each dimension share the same kernel parameters.

Any idea on how to do this? I was thinking of making custom GP models with that option in mind, but wanted to ask here first if the current GP implementations can do it.

Hi @luis, currently, we don’t have support for batch GPs. I would recommend using GPyTorch instead.