Cannot import because of tensorboard error, yet tensorboard is installed about pytorch-ts

[kashif]

right this warning is coming from pytorch... in any case can you kindly try to update:

pip install -U tensorboard tensorboard-plugin-wit

and try again? Thanks!

[AlexMRuch]

I think this has to do with a pip vs. conda install. Your pip install defaults to CPU-only pytorch (why not detect cuda presence and do something like pip install torch==#.#.#+cu###), and I tend to use conda to install DL libraries. Do you all have a conda yaml up yet for conda-forge?

[AlexMRuch]

Yeah, trying to fix it now with a pip install of the above. Gosh, pip and conda are always such a labyrinth even with brand new envs...

[AlexMRuch]

Okay, using the pip version for everything across the board works. So a break happens if you have conda pytorch and not pip pytroch. So I guess if one wants GPUs for PyTorchTS they should make sure to 1) create a new conda env, 2) conda install pip, 2) pip install their appropriate GPU version of PyTorch, 3) and finally pip install pytorchts. Hope that helps anyone else hitting this!

[kashif]

thanks @AlexMRuch I appreciate you using this library.

[AlexMRuch]

Oh, I super appreciate you all making this happen, @kashif! There are so many DL time series models out there but it doesn't seem like many people are really integrating all those into an abstract and generalizable library. Super happy to have stumbled upon it!