Make pip installable

[MilesCranmer]

This enables you to install it with:

pip install git+https://github.com/ldeecke/gmm-torch

rather than cloning and copying the code into a codebase

[ldeecke]

Makes sense and looks good in principle — thanks for the PR! 👍

Some suggestions:

• could rename from gmm_torch/gmm.py to gmm/model.py
• {example,test.py} seem to loose their functionality
• what is the rationale for dropping torch from requirements?

[MilesCranmer]

Regarding dropping torch, I’ve noticed it will often mess up installations of torch when you put it in setup. For example, if you’ve installed the GPU version, the setup.py might make pip switch it to the CPU version. So it’s better to assume someone will install pytorch manually using the steps on https://pytorch.org/get-started/locally/ which vary by platform and accelerator.

[Linux-cpp-lisp]

It would be great to have this available in PyPI; it would make it much easier to use and ensure that downstream packages are getting any updates or bugfixes. It's possible (and really quite easy) to set up GitHub actions to upload to PyPI every time you make a release in GitHub, see for example in one of my packages here: https://github.com/Linux-cpp-lisp/opt_einsum_fx/blob/main/.github/workflows/release.yml

Also just seconding @MilesCranmer 's point about torch as a dependency... I've never seen pip do anything sane or helpful with it, but it's definitely ruined a lot of environments and wasted a lot of bandwidth.

[AndreasGerken]

I would also love to have this as pip installable. It's very easy and saves each users setup time. If it would be in PyPi it would be even more amazing and it would probably find even more users.

By now there is no specific torch-gpu version anymore (was there one once?) so there should not be any fear of messing up environments.