bayespy.nodes.Gaussian.lower_bound_contribution

Gaussian.lower_bound_contribution(gradient=False, ignore_masked=True)

Compute E[ log p(X|parents) - log q(X) ]

If deterministic annealing is used, the term E[ -log q(X) ] is divided by the anneling coefficient. That is, phi and cgf of q are multiplied by the temperature (inverse annealing coefficient).