Documentation for paragami

Parameter folding and flattening, parameter origami [1]: paragami!

This is a library (very much still in development) intended to make sensitivity analysis easier for optimization problems. The core functionality consists of tools for “folding” and “flattening” collections of parameters – i.e., for converting data structures of constrained parameters to and from vectors of unconstrained parameters.

The purpose is to automate much of the boilerplate required to perform optimization and sensitivity analysis for statistical problems that employ optimization or estimating equations.

The functionality of paragami can be divided into three mutually supportive pieces:

  • Tools for converting structured parameters to and from “flattened” representations,
  • Tools for wrapping functions to accept flattened parameters as arguments, and
  • Tools for using functions that accept flattened parameters to perform sensitivity analysis.

A good place to get started is the Examples.

For additional background and motivation, see the following papers:

Covariances, Robustness, and Variational Bayes
Ryan Giordano, Tamara Broderick, Michael I. Jordan

A Swiss Army Infinitesimal Jackknife
Ryan Giordano, Will Stephenson, Runjing Liu, Michael I. Jordan, Tamara Broderick

Evaluating Sensitivity to the Stick Breaking Prior in Bayesian Nonparametrics
Runjing Liu, Ryan Giordano, Michael I. Jordan, Tamara Broderick
[1]Thanks to Stéfan van der Walt for the suggesting the package name.