[QUESTION] Is there a recommended way to use darts with cpu version of pytorch ?

[trya2l]

Hello,

I have this use case where I need to use darts with PyTorch, but I do not have access/need for a GPU.

I was able to install darts with PyTorch CPU version with

pip install u8darts 
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install pytorch_lightning

Is there a better or more official way to install Darts with CPU-only PyTorch? For example, could there be an option like pip install u8darts[torch-cpu] similar to existing extras ?

If so, would it be appropriate to add these instructions to the official installation docs here?

https://github.com/unit8co/darts/blob/master/INSTALL.md

Thanks!

[madtoinou]

According to the various resources I could find online, what you described seems to be a nice way to install Darts with the CPU-only version of PyTorch. I would maybe just install PyTorch before installing Darts, so that it's not installed twice.

At the moment, there is no official way/flavor of Darts with PyTorch for CPU but we might consider it.

[dennisbader]

@madtoinou u8darts comes without torch :)

You could simplify it a bit with:

pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install u8darts[torch]

This would install all our torch dependencies though:

pytorch-lightning>=1.5.0
tensorboardX>=2.1
torch>=1.8.0

Since torch would already installed in the CPU version, it should not install the GPU version.

If you don't need tensorboardX then you can use what you suggested initially :)

[trya2l]

Hello @dennisbader @madtoinou,

Thanks for the tips. I've taken the time to try, and I can confirm it works as intended.
Would you find it appropriate if I sent a PR to add this use case to the docs ?

darts/INSTALL.md

Lines 1 to 18 in bd53fe0

	# Installation Guide
	Below, we detail how to install Darts using either `conda` or `pip`.
	## From PyPI
	Install Darts with all models except the ones from optional dependencies (Prophet, LightGBM, CatBoost, see more on that [here](#enabling-optional-dependencies)): `pip install darts`.
	If this fails on your platform, please follow the official installation
	guide for [PyTorch](https://pytorch.org/get-started/locally/), then try installing Darts again.
	As some dependencies are relatively big or involve non-Python dependencies,
	we also maintain the `u8darts` package, which provides the following alternate lighter install options:
	* Install Darts with all available models: `pip install "u8darts[all]"`
	* Install core only (without neural networks, Prophet, LightGBM and Catboost): `pip install u8darts`
	* Install core + Prophet + LightGBM + CatBoost: `pip install "u8darts[notorch]"`
	* Install core + neural networks (PyTorch): `pip install "u8darts[torch]"` (equivalent to `pip install darts`)

After these lines I would add an item to the list in the form of

• Install core + neural networks CPU only (using PyTorch CPU wheel): first install torch pip install torch --index-url https://download.pytorch.org/whl/cpu then u8darts pip install "u8darts[torch]"