iterLap: Approximate probability densities by iterated Laplace
Approximations
The iterLap (iterated Laplace approximation) algorithm
approximates a general (possibly non-normalized) probability
density on R^p, by repeated Laplace approximations to the
difference between current approximation and true density (on
log scale). The final approximation is a mixture of
multivariate normal distributions and might be used for example
as a proposal distribution for importance sampling (eg in
Bayesian applications). The algorithm can be seen as a
computational generalization of the Laplace approximation
suitable for skew or multimodal densities.
Downloads: