Changelog

v0.16.0 (August 2024)

Bumped Version: Updated version and README.
CI Updates: Separated linting and testing actions.
Documentation: Updated docs and improved handling of hidden features.
Bug Fixes:
- Fixed docstrings and embedding-net handling messages.
- Resolved issues with batched sampling and associated tests (#35).
- Fixed CI configurations and Pyright errors.
- Corrected version retrieval process.
Enhancements:
- Added contribution guide (#5) and code of conduct.
- Replaced setup with pyproject.toml.
- Refactored CI; added pre-commit hooks and formatting.
- Extended tests (#32).
License Change: Switched to Apache 2.0 (#28).
Miscellaneous: Performed Ruff checks and automatic fixes.

v0.15.2

Replace torch.triangular_solve() with torch.linalg.solve_triangular (#26).

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)

v0.14.2

fix device bug in analytical MoG sampling (#19).

v0.14.1

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

v0.14.0

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

v0.13.1

Fix GPU bug in MDN.

v0.13.0

Upgrade dependency on nflows to 0.13 to allow GPU support.

v0.12.0

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)