How to choose q() to sample theta? Usally gausian, for the parameter u and sigma, we can get from cross validation.
Importance sampling can only work on low dimension for Theta. And it q must support posterior in most of area.
Learned from CPSC 540, machine learning