"Parameter origami" -- folding and unfolding collections of parameters for optimization and sensitivity analysis.
In order to use out-of-the-box analysis tools (e.g. optimization, sensitivity analysis, Hamiltonian Monte Carlo), data analysts typically have to convert structured, constrained parameters (such as covariance matrices) to and from unconstrained one-dimensional vectors. The goal of paragami is to automate much of the boilerplate required for such conversions, all in a way that is compatible with general-purpose automatic differentiation. The core functionality consists of tools for “folding” and “flattening” collections of parameters – i.e., for converting structured data structures to and from 1-d vector representations that can legally take any value and still represent valid structured parameters.
https://github.com/rgiordan/paragami, or just $ pip install paragami
