v2.0.1

What's Changed

• BUG: fix exception not raised bug in python by @zoj613 in #29
• MAINT/REL: modernize build and support 3.8-3.11 by @zoj613 in #31
• REL: Ensure openblas and lapack installation in release workflow. by @zoj613 in #32
• BLD: Skip building of musllinux wheels by @zoj613 in #33
• REL: Update pypi uploading workflow by @zoj613 in #34

Full Changelog: v2.0.0...v2.0.1

v2.0.0

• a_type and o_type parameters of structured_precision_mvnorm are now strings instead of numbers. This change is not backwards compatible.
• support for python 3.6 has been dropped.

v1.0.0

• Python API has been re-designed to be a functional interface instead of being class-based. This is to help make the code easier to plugin into other libraries.
• The API is also Numpy/scipy friendly and allows the user to make use of the bitgenerators offered by numpy's numpy.random.Generator via the optional keyword argument random_state.
• The need for cblas.h and lapacke.h headers has been relaxed. Only an installation of LAPACK is a requirement for using the C library.
• The new python API is not backwards compatible.

v0.2.0

• structured precision sampler has been refactored and both memory and speed efficiency improved.
• RNG C code has been refactored.
• If the out parameter is passed to the python functions, then no output will be explicitly returned.
• Fix error with input dimension check when a low rand phi is used.
• A faster random sampling method for the standard normal distribution has been added (Ziggurat algorithm)

v0.1.1

• Fixed a bug that results when sampling using a G matrix with only one row.

First release

First release to PyPi.