v2.0.1

What's Changed

• BUILD: Add support for Python 3.13 and drop 3.9. by @zoj613 in #128

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

v2.0.0

What's Changed

• BUILD: Remove cython as hard build dependency. by @zoj613 in #125
• BUILD, MAINT: Migrate to Cython 3 and use the Iterator API instead of PyArray_MultiIter_* functions for iteration. by @zoj613 in #126
• BUILD: Add support for numpy 2.0 by @zoj613 in #127

Full Changelog: v1.3.6...v2.0.0

v1.3.6

What's Changed

• DEV: Remove codecov from requirements file. by @zoj613 in #122
• BUILD: Add support for python 3.12 and drop 3.8. by @zoj613 in #123
• REL: Prepare for v1.3.6 release. by @zoj613 in #124

Full Changelog: v1.3.5...v1.3.6

v1.3.5

What's Changed

• MAINT: Add support for python 3.11 and drop support for 3.7 by @zoj613 in #121

Full Changelog: v1.3.4...v1.3.5

v1.3.4

• Ensure full precision for 32 bit floating point random values (see #117 )
• Return NaN for out of range parameter values when disable_checks=True (see #116 )

v1.3.3

• Adds support for python 3.10 and drops 3.6

v1.3.2

• Fixes a type casting bug that occurs when passing the scale parameter h as a numpy array when the method is explicitly set to devroye. See #85.
• Fixes a bug with the devroye method hanging unexpectedly in some corner cases. Details at #83
• Non-scalar arguments for h and z can now be broadcasted together with a specified size value. This is consistent more with NumPy's random functions.
• Adds support for Windows OS and adds wheels for both MacOS X and Windows.
• Adds installation of the package via conda-forge for users who rely on conda to install packages.
• the -march=native compiler option is removed when building the package to avoid problems with conda distributions of the package.

v1.3.2-beta.3

• Improves support for Windows build. It should build successfully on windows sever >=2016

pip install polyagamma==1.3.2b3

v1.3.2-beta.2

• Adds support for Windows 10 and adds wheels for MacOS X and Windows.

pip install polyagamma==1.3.2b2

v1.3.2-beta.1

• Non-scalar arguments for h and z can now be broadcasted together with a specified size value. This is consistent more with NumPy's random functions.