numba_quaternion

Quaternion operations that can be used within Numba-jit functions.

Introduction

This package contains some numba-jit-compiled functions that perform Quaternion operations and a convenient class Quaternion that provide convenient methods wrapping around those functions.

Quaternion behaves like a Numpy array containing quaternion, e.g. respect Numpy broadcast operations, but without really imitating a numpy.ndarray and implemented a dtype.

This design allows you to write any jit-compiled functions involving those provided jit-compiled functions, and then write your own class methods that calls those functions as a convenient interface (by class inheritance.)

If you do not care to use Quaternion in other jit-compiled functions you write, check out packages below instead.

Installation

pip install numba-quaternion
# or
conda install -c conda-forge numba_quaternion

Other Python quaternion projects

Other Python projects that implements quaternions and I knew and used are:

• zonca/quaternionarray: written in pure Python using Numpy. Note that unusually they put the real part in the last column. lastcol_quat_to_canonical and canonical_quat_to_lastcol convert between those and the canonical ordering (where real part comes first.)

• hpc4cmb/toast: toast.qarray is a reimplementation of the above quaternionarray package in C++ with the same interface, and following the same convention.

• moble/quaternion: implement Quaternion as a Numpy dtype in C.

• moble/quaternionic: implement Quaternion as a Numpy dtype using Numba. This package is inspired by my expectation of quaternionic—I expected I could use them in a Numba-jit-compiled function but it doesn’t.