created_at2023-06-03 15:50:26.289211
updated_at2023-06-03 15:50:26.289211
descriptionTools for statistical unit testing of sampling procedures



License: MIT build

test-sampler: Unit Test Tool for Sampling Algorithms

The package is intended to provide utilities to unit test sampling algorithms.

In particular, when developing Monte Carlo particle transport codes, one needs to write procedures to sample a large number of complex distributions that describe different physical interactions (various types of scattering, fissions etc.). The models are usually well described by differential cross-sections $\frac{d\sigma}{dE'd\Omega}(E)$ for which explicit expressions exist. However, in practice getting the normalisation to convert cross-section into the probability density function (pdf) may require tricky numerical integration. Thus, in short, getting the shape of pdf is often easy, getting the pdf or cumulative distribution function (cdf) is not.

This package provides two components to enable unit testing of sampling algorithms.

  • An universal sampler which can draw samples from an arbitrary shape function that describes the continuous 1D distribution on a bounded support.
  • A selection of two-sample statistical tests to verify that two samples were drawn from the same distribution.

Thus one can use the universal sampler to (inefficiently) generate reference set of samples from a shape function and compare them against the 'production' algorithm.

Compile docs

The package is using Katex to render equations as the result to compile documentation locally extra flags need to be provided to cargo:

export RUSTDOCFLAGS="--html-in-header <path-to-repo>/katex-header.html"

To compile docs from the root of the repositry run::

RUSTDOCFLAGS="--html-in-header katex-header.html" cargo doc --no-deps
Commit count: 20

cargo fmt