Implementation and derivation of "Variational Bayesian inference for a nonlinear forward model." [Chappell et al. 2008] for arbitrary, user-defined model errors.
@aradermacher added missing constant (w.r.t the optimization) prior terms such that the VB free energy actually expresses the log evidence -- given that the KL(true posterior, MVN posterior) == 0.
EDIT: ~With those missing terms present, I added the evidence.py file (not yet a test) that compares the vb free energy to the log evidence from a dynesty sample. This sampling is too slow for a test file and I'm sure that we can come up with an analytic formula based on https://www.econstor.eu/bitstream/10419/85883/1/02084.pdf (eq 3)~
Closes #54
@aradermacher added missing constant (w.r.t the optimization) prior terms such that the VB free energy actually expresses the log evidence -- given that the KL(true posterior, MVN posterior) == 0.
EDIT: ~With those missing terms present, I added the
evidence.pyfile (not yet a test) that compares the vb free energy to the log evidence from a dynesty sample. This sampling is too slow for a test file and I'm sure that we can come up with an analytic formula based on https://www.econstor.eu/bitstream/10419/85883/1/02084.pdf (eq 3)~Actually did that now.