recovering latent GP in LMC MOGP

[michaelcao28]

Hi there,

I am wondering if it is possible to obtain the latent GP in an LMC MOGP model, like the model in this tutorial. In this tutorial, there were 3 latent GPs. Is there a way to retrieve the latent GP? Any tips would be much appreciated!

Thanks!

[gpleiss]

At the moment, there's not really a good (non-hacky) way to do this. It would make sense for there to be an argument that could get passed into LMCVariationalStrategy.call. Something like:

model(x, latent_functions=True)  # latent_functions gets passed into LMCVariationalStrategy.__call__

Please feel free to open up a feature request for this, but I'm pretty backlogged so I probably won't get to it for a while...