One of the biggest pain of PyMC GP is that it holds reference to live variables from the graph in the class instance (self.sigma or whatever), so we can't operate on the PyMC model as a graph directly (say pm.observe, pm.do, pm.marginalize ...).
Does this sort of problem persist in this design?
One of the biggest pain of PyMC GP is that it holds reference to live variables from the graph in the class instance (self.sigma or whatever), so we can't operate on the PyMC model as a graph directly (say pm.observe, pm.do, pm.marginalize ...).
Does this sort of problem persist in this design?