StanPathfinder.jl

Project Status	Build Status

Purpose

StanPathfinder.jl wraps cmdstan's pathfinder method. It is documented in Zhang, Lu, Bob Carpenter, Andrew Gelman, and Aki Vehtari (2022): “Pathfinder: Parallel Quasi-Newton Variational Inference.” Journal of Machine Learning Research 23 (306): 1–49. See also Stan's cmdstan manual (chapter 6) for details.

An example can be found here and a Pluto notebook example can be found here.

Note: This is an initial implementation. A few changes are to be expected!

Installation

This package is registered, you can install it with

pkg> add StanPathfinder.jl

You need a working cmdstan installation, the path of which you should specify either in CMDSTAN or JULIA_CMDSTAN_HOME, eg in your ~/.julia/config/startup.jl have a line like

# CmdStan setup
ENV["CMDSTAN"] = expanduser("~/src/cmdstan-2.35.0/") # replace with your path

Usage

It is recommended that you start your Julia process with multiple worker processes to take advantage of parallel sampling, eg

julia -p auto

Otherwise, stan_sample will use a single process.

Use this package like this:

using StanPathfinder

See the docstrings (in particular ?StanPathfinder) for more.