We have added support for new machine learning models, which learn the target distribution from a small subset of posterior samples. In this case we consider normalising flows (Real-NVP and Rational Quadratic Splines) though in later releases one could expand this to a larger set of models. This new functionality may be installed directly from pypi on macOS apple silicon devices and manylinux using `pip install harmonic`.
What's Changed
* add normalising flows as target models (Real-NVP and Rational Quadratic Splines)
* Retire existing classical models (e.g. hypersphere etc.) to model_legacy. Of course, all functionality is still available in this new location.
New Contributors
* alicjapolanska wrote and implemented these additions to Harmonic, reviewed by CosmoMatt and jasonmcewen.
**Full Changelog**: https://github.com/astro-informatics/harmonic/compare/v1.1.1...v1.2.0