ParetoSmoothedImportanceSampling.jl

Project Status	Build Status

Purpose of this package

This package implements model comparison methods as used and explained in StatisticalRethinking (chapter 7). Thus, ParetoSmoothedImportanceSampling.jl is part of the StatisticalRethinking family of packages.

The most important methods are Pareto smoothed importance sampling (PSIS) and PSIS leave-one-out cross-validation based on the Matlab package called PSIS by Aki Vehtari. The Julia translation has been done by @alvaro1101 (on Github) in a (unpublished) package called PSIS.jl.

Updates for Julia v1+, the new Pkg ecosystem and the addition of WAIC and pk utilities have been done by Rob J Goedman.

A new package, ParetoSmooth.jl, might over time replace the inner parts of this package.

Installation

ParetoSmoothedImportanceSampling.jl can be installed with:

Pkg.add("ParetoSmoothedImportanceSampling")

Usually I have only a few packages permanently installed, e.g.:

(@v1.6) pkg> st
      Status `~/.julia/environments/v1.6/Project.toml`
  [634d3b9d] DrWatson v1.16.6
  [44cfe95a] Pkg

To use the demonstration Pluto notebooks, you can add:

  [c3e4b0f8] Pluto v0.12.18
  [7f904dfe] PlutoUI v0.6.11

To run the notebooks, I typically use an alias:

alias pluto="clear; j -i -e 'using Pkg; import Pluto; Pluto.run()'"

and then do:

$ cd ~/.julia/dev/ParetoSmoothedImportanceSampling
$ pluto

to start Pluto from within that directory.

The cars WAIC example requires RDatasets.jl to be installed and functioning.

Included functions

psisloo() - Pareto smoothed importance sampling leave-one-out log predictive densities.

psislw() - Pareto smoothed importance sampling.

waic() - Compute WAIC for a loglikelihood matrix.

dic() - Deviance Information Criterion.

pk_qualify() - Show location of pk values.

pk_plot() - Plot pk values.

Additional function:

gpdfitnew() - Estimate the paramaters for the Generalized Pareto Distribution (GPD).

gpinv() - Inverse Generalised Pareto distribution function.

var2() - Uncorrected variance.

Corresponding R code

Corresponding R code for the PSIS methods can be found in R package called loo which is available in CRAN.

References

• Aki Vehtari, Andrew Gelman and Jonah Gabry (2016). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Statistics and Computing, doi:10.1007/s11222-016-9696-4. arXiv preprint arXiv:1507.04544
• Aki Vehtari, Andrew Gelman and Jonah Gabry (2016). Pareto smoothed importance sampling. arXiv preprint arXiv:1507.02646
• Jin Zhang & Michael A. Stephens (2009) A New and Efficient Estimation Method for the Generalized Pareto Distribution, Technometrics, 51:3, 316-325, DOI: 10.1198/tech.2009.08017