search

mehta-lab / waveorder

Wave optical models and inverse algorithms for label-agnostic imaging of density & orientation.
BSD 3-Clause "New" or "Revised" License
12 stars 3 forks source link

Make `torch` optional dependency? #130

Closed ieivanov closed 12 months ago

ieivanov commented 12 months ago

Does it make sense to make torch an optional dependency? How easy it it to write non-torch accelerated versions of the waveorder algorithms? Are there cases where we'd like to use waveorder but cannon install torch, e.g. no GPU?

mattersoflight commented 12 months ago

torch is a generic replacement for numpy on CPU, GPU, or integrated chips like M1.

talonchandler commented 12 months ago

@ieivanov what's the motivation here? Have you run into issues using/installing torch?

So far I haven't had any trouble with pip install torch on machines with or without gpus.

ieivanov commented 12 months ago

No issues, just asking for now. I'll try installing torch on mantis and I'll let you know if I run into any problems.

ieivanov commented 12 months ago

No issues installing torch on mantis