How can I install torch_geometric in remote server without manually do it.

[Anak2016]

📚 Installation

How can I install torch_geometric in remote server without manually do it.
I have no choice now but to use cloud to run my program and the problem I always have with torch_geometric is that it can't simply be install by "pip install torch_geometric" but i have to also install torch_sparse, ... etc which depends on cuda version and pytorch version.

I usually have to do it manually everytime, but now I am using third party cloud and have no choice to manually install those packages.

Is there a way to install torch_geometric without manually do it?

it is probably something super easy, but like other programming problem if I don't know the word to search for the problem, no way I can find solution unless I have to read the a whole books on similar topic.

[rusty1s]

Are you speaking about a requirements file? I'm not really sure by what you mean by you cannot automate it. As long as your PyTorch/CUDA version is known, you can easily install it via pip. Can you give a concrete example which shows the problem of installing PyG?

[Anak2016]

May be I just don't know how to do it correctly.

I tried using both pip freeze and conda list to produce requirements file.

but from what I know is that torch_geometric have to install based on specified version of CUDA and python. Wil using pip freeze or conda list allow for this flexibility where the neccesary will be install automatically based on verion of CUDA and python? because when I run it in Floydhub, a cloud service. The machine fail to install torch_geometric.

Note: I don't know cuda version of floydhub machine. May be there is a way to get this information, I haven't look in to it.

Thank you

[Awannaphasch2016]

@rusty1s Could you please follow up with this matter? i am the same person as Anak2016.

[rusty1s]

Sorry, I was offline for a few days.
In general, you need to know the CUDA version installed to install PyTorch Geometric. You can find it by running nvcc --version or torch.version.cuda.
Then you can create a requirement file requirements.txtlike this:

-f https://pytorch-geometric.com/whl/torch-1.5.0.html
torch-scatter==latest+${CUDA}
torch-sparse==latest+${CUDA}
torch-geometric

and install it via pip install -r requirements.txt