Motivation

2-point statistics are important for characterizing the properties of turbulence (2-point statistics comes up in other contexts like cosmology). There hasn’t been an easy-to-use package for computing these quantities, until now.

The pairstat package is most useful for datasets where Fourier methods are problematic (e.g. you don’t have a regularly spaced periodic grid). Before developing pairstat, I performed similar calculations by processing the outputs of scipy.spatial.distance.pdist and scipy.spatial.distance.cdist functions. This package implements equivalent functionality that uses more specialized C++ code in order to perform the calculation faster and with far less memory. [1] It also supports parallelization (more on that below).

Installation

As long as you have a C++ compiler, the easiest way to get the package is by invoking

$ python -m pip install pairstat

The package is automatically compiled with OpenMP support if the compiler supports it. To confirm that pairstat was compiled with OpenMP support, you can check whether the output from the following command mentions OpenMP:

$ python -m pairstat

See our Installation Guide for more details (especially if the package wasn’t compiled with OpenMP support).

Key-Features: Parallelism and Scalability

The key feature of this package is the support for parallelism. If a compatible compiler is used to build this package, it will automatically be built with OpenMP support for parallelizing calculations of structure functions and correlation functions.

Undocumented machinery also exists to help use this functionality to parallelize calculations across machines on a computing cluster (e.g. with MPI). We plan to document this machinery in the near future.

The other important feature, is memory usage. The memory usage is independent of the number of points. A naive implementation of equivalent calculation using scipy functionality has memory usage that scales with the number of pairs of points (i.e. the number of points squared for auto-correlation). In other words, this function is far more scalable that the alternative.

Current Status

We are planning to replace the C++ and Cython logic with the rust logic before the 1.0 release. This rewrite will allow us to significantly improve the code quality.

Contributions and Feature requests are welcome!

License

pairstat is dual-licensed under either the MIT license and the Apache License (Version 2.0).

Footnotes