Open max-gains opened 1 year ago
This example uses only a scaled RBF kernel (not a multi-output kernel) and a MultivariateNormal dist, not a MultitaskMultivariateNormal.
This is actually correct. The DeepGP outputs multiple independent MultitaskNormal distributions (based on the width of the last layer) which are then combined into a MultitaskMultivariateNormal.
There are also differences between the code and the supporting writing (which says a MultitaskMultivariateNormal should be used).
I understand that the docs might be confusing, so if you have suggestions for how to make them clearer we'd be open to a fix!
📚 Documentation/Examples
I think the docs for deep multi-output regression are wrong: https://docs.gpytorch.ai/en/stable/examples/05_Deep_Gaussian_Processes/DGP_Multitask_Regression.html
This example uses only a scaled RBF kernel (not a multi-output kernel) and a MultivariateNormal dist, not a MultitaskMultivariateNormal. There are also differences between the code and the supporting writing (which says a MultitaskMultivariateNormal should be used). I can provide a fix if requested.