Formulating a hierarchical multivariate Gaussian process using the Hilbert space approximation

[avehtari]

@jr-leary7

Does the above approach make sense as a means of estimating a hierarchical approximate GP ? If not, in what way have I misspecified the model / assumptions / data ?

It looks like you have the magnitude of each GP twice as you have sigma_alpha and amplitude.

See also discussion in Hilbert space Gaussian process for multiple time series

Unfortunately, right now I don’t have time to examine your code in more detail.