search

bayesiains / nflows

Normalizing flows in PyTorch
MIT License
853 stars 119 forks source link

Consider packaging and distribution on conda-forge #82

Open matthewfeickert opened 6 hours ago

matthewfeickert commented 6 hours ago

In addition to publishing distributions on PyPI, it would be very nice to also distribute on conda-forge and as nflows is a pure-Python package that would be technically easy.

However, given the required dependencies

https://github.com/bayesiains/nflows/blob/3b122e5bbc14ed196301969c12d1c2d94fdfba47/setup.py#L20-L27

umnn is not yet on conda-forge and would be required to be distributed there first.

Related issues:

matthewfeickert commented 6 hours ago

Ah, seems that https://github.com/AWehenkel/UMNN is unmaintained research software so getting https://pypi.org/project/UMNN/ on conda-forge might take some additional work. There might be hope though given that there were releases on PyPI as new as a year ago (December, 2023). Ping @AWehenkel in the event they can offer help (also hello again!).