I was curious for general jacobian adjustments usecases and this was also helpful. So maybe
needed:
- priors that go beyond
parameter
block - jacobian of reparameterization is nonconstant (eg. nonlinear transformation)
- jacobian of reparameterization is constant but Bayes factor is needed
not needed:
- priors restricted on
parameter
block - jacobian of reparameterization is constant and only inference is needed