@essicolo and @DanhPhan, you can indeed use PyMC for multi-output GPs. Basically you just need to add an extra input indicating which output you’re observing/predicting. See the small example here: Coregionalization model for two separable multidimensional Gaussian Process - #4 by bwengals. I’ve built it into my package for building GPs from tabular data, you can see it in action here and the code here.
2 Likes