v0.16.0

What's Changed

• Mdn cleanup/deprecating hidden features by @anasbekheit in #27
• extend tests. by @janfb in #32
• Update CI; add ruff; add pyproject.toml by @janfb in #33
• fix #5: add contribution guide by @janfb in #34
• change license to Apache 2.0 by @janfb in #28
• fix #35: fix batched MDN sampling, fix tests. by @janfb in #36
• bump version, update readme by @janfb in #37

New Contributors

• @anasbekheit made their first contribution in #27

Full Changelog: v0.15.2...v0.16.0

v0.15.1

• Fixes to version v0.15.0, in which the release did not contain the desired fixes.

v0.15.0

• epsilon is added after multiplication of Cholesky-factors for better numerical stability (#22)

Refactor to allow analytic MoG evaluation

refactor MoG to allow evaluation with analytically determined parameters (#17).

Upgrade to nflows 0.14.

Upgrade dependency on nflows to 0.14 for NSF fix (#25).

Bug fix for GPU support

Small but essential fix for inferring the correct device during prelocation.

Upgrade to nflows 0.13

• upgrade to new nflows 0.13 which now allows NN training on the GPU.

Upgrade to nflows 0.12

• Upgraded dependency on nflows to 0.12; explicitly allow Python 3.6 (#13)
• Support for batch_x with only one feature in mixture density networks (thanks to
  Harry Fu, #12)
• Imports from nflows.transforms.splines now available (thanks to Miles Cranmer, #11)