RELEASE requirements?

Hi,

When doing:

conda install pytorch torchvision -c pytorch
pip install -i https://test.pypi.org/simple/ kymatio

then cloning the corresponding release, and doing pytest, I have tons of errors that are evidences that none of the requirements.txt packages are installed. Actually, this makes me think we should have a pip for the GPU version(like pip install kymatio-gpu)?

https://github.com/kymatio/kymatio/blob/master/requirements.txt

[edouardoyallon]

ok. So I’ll edit such that:

pip install scikit-cuda cupy

replaces the requirements_cuda.txt

[janden]

I would be in favor of a pip install kymatio-gpu, but I can see how that is a commitment to have two, rather than one, PyPI hosted mirrors.

Yeah I’d rather go that route as well. It’s a little more work for us, but the users’ lives would be so much simpler. The question is whether we’d want to go PyTorch all the way and have “kymatio” and “kymatio-cpu”?