cdriveraus/stanoptimis: Optimization and importance sampling for Stan models.

A small package containing a 3 step routine for fitting Stan models, and limited diagnostics. For complex models with low to moderate numbers of parameters, where the maximum a-posteriori estimate provides a useful starting point, this approach 'may' allow for faster sampling from the posterior. The optional step 1 uses differential evolution, linking to the DEoptim package. Step 2 then uses a BFGS optimizer from the mize package. Step 3 computes a Hessian, or approximation of the Hessian, and uses this either for: a) directly sampling from, for fast but inaccurate representation of the posterior. b) as the basis for an initial target distribution for an adaptive importance sampling procedure.

install.packages("remotes")
remotes::install_github("cdriveraus/stanoptimis")