Currently gp.Latent can only handle a single GP observation, so f ~ GP(0, k(x, x')). It should be generalized to allow f_i ~ GP(0, k(x, x')), which is multiple observations of the same GP.
It "should" be pretty straightforward to extend all the GP implementations to handle this case, but it's probably most important for gp.Latent.
Context for the issue:
Currently
gp.Latent
can only handle a single GP observation, sof ~ GP(0, k(x, x'))
. It should be generalized to allowf_i ~ GP(0, k(x, x'))
, which is multiple observations of the same GP.It "should" be pretty straightforward to extend all the GP implementations to handle this case, but it's probably most important for
gp.Latent
.