Laplace approximation sampler command

[bob-carpenter]

Summary:

Add a method to perform sampling from a Laplace approximation.

Description:

Add a Laplace approximation sampler that returns a sample .csv file with:

• standard output configuration as comments at the top
• header for names of constrained parameters, lp__ (target log density), and lq__ (approximation log density)
• rows corresponding to draws on the constrained scale

The arguments required are

• mode: a specification of a mode on the constrained scale using either

  • .csv format: a .csv file containing constrained parameters
    • # initial: comments that will be ignored; these can be anywhere
    • header: comma-separated names of constrained parameters in same format as our other output (e.g., sampling or optimization), with any names ending in double underscores (__) ignored; this must come before any value lines
    • values: there can be any number of rows for values, but only the final line is used
  • .json format: json specification of unconstrained values in the same format as used for initialization in other methods; all parameter values must be specified, and non-parameter values will be ignored

• jacobian: boolean value indicating whether to include Jacobian adjustment or not; including provides max a posteriori (MAP) estimate, not including produces a (penalized) maximum likelihood estimate (MLE); default is true

• draws: a positive integer indicating number of approximate draws to return; default is 100

• refresh: an integer indicating how often to provide messages signaling which iteration has been processed; if this is not positive (<= 0), then no output will be produced for iterations (and ideally not for other messages from model class); default is 100

• seed: a short unsigned integer to use as the random number generator; default is to generate randomly based on time

Additional Information:

The service function implementing Laplace sampling is: https://github.com/stan-dev/stan/blob/develop/src/stan/services/optimize/laplace_sample.hpp

Current Version:

v2.30.1