Wesley Maddox

Something like this ought to work for you (and I probably ought to clean this up as a PR at some point): ```python # using botorch.models.KroneckerMultitaskGP here but the API...

> I did not find a clear example but I guess ScaleKernel is the one to use to include spatial dimension. I essentially want to predict t+1 from t. Output...

This is actually not quite a bug but seems to be caused by your usage of `fast_pred_var` in the inference step which isn't strictly necessary here. If I remove that...

That was mostly a tool for debugging what's going on, you ought to be able to use either here. In general, it depends on if you want the posterior over...

Yes, that's the correct understanding.

Couldn't this be done by reshaping the two output dimensions and adding in an output flag to your inputs describing which output they come from? Your kernel would then probably...

Easiest fix is probably just to perform that computation on the CPU. There's probably minimal use-cases where the inter-task covariance matrix is actually large enough to really be slowed down...

I also verified that this issue is somehow fixed on pytorch nightly.

So the strategy proposed in the matern graph kernel paper is to do an interdomain inducing point strategy where the inducing points are the eigenvalues / vectors of the covariance...

Can you explain in a bit more detail? It doesn't look like the proximal issue of an explicit matrix being formed in the forwards pass has been resolved (still [here](https://github.com/cornellius-gp/gpytorch/blob/2e0bbc9f59e4b4b54780c3e55db784c3d2c9a5bf/gpytorch/kernels/spectral_mixture_kernel.py#L335)....