Is there a way to have ‘latent’ factors in LCM multi output GPs? Following from the baseball example, there are 5 players that they all influence each other but is there a way to say that within those five players only n (where n<5) subgroups should really exist? Current implementation of W in pm.gp.cov.Coregion wants to have as many W matrices as GP outputs but is there a way to have a model that has N outputs (the 5 players) but only 2 (or any n<5) subgroups? It’s not obvious from the code that playing around with the shape of the W matrix is the way to do this but would be really great to have multi output GPs with a lower rank approach to them.