Open fhui28 opened 3 years ago
Also adding uncertainty of G and Sigma. This may be remotely possible since we could calculate the REML/ML wrt with G and Sigma via TMB and get the hessian off of that.
Alternatively, bootstrap to get a poor person's solution?
Add parametric structures for Sigma e.g., exponential/Matern covariance with distances dependent on time. This will substantially reduce the number of parameters to estimate, but also according to AZM this tends to work better for temporal forecasting compared to straight up using temporal basis functions...
Enough said! Don't really know how to solve this..more like things to think about