a question laplace's approximation:
in laplace's method, need find mode of function , take second order taylor's expansion. first order term vanish (since gradient 0 @ local optimum), , second order term used give covariance of gaussian. wondering if doing mcmc , "point" found not local optimum (a.k.a gradient not zero), influence covariance matrix? matrix non-positive-definite ?
Comments
Post a Comment