Installation

TorchDrug can be installed on either Linux, Windows or macOS. It is compatible with 3.7 <= Python <= 3.10 and PyTorch >= 1.8.0.

For Windows

From Conda

conda install torchdrug -c milagraph -c conda-forge -c pytorch -c pyg

From Pip

First, let’s install PyTorch.

pip install torch

To install torch-scatter, we need to check the version of PyTorch and CUDA.

We can get the version of PyTorch by python3 -c "import torch; print(torch.__version__"). The version of CUDA can be get by nvcc -V. For example, if our PyTorch is 1.8.0 and CUDA is 10.2, the command should be

pip install torch-scatter torch-cluster -f https://pytorch-geometric.com/whl/torch-1.8.0+cu102.html

Replace the versions in the above url according to your case. If you don’t have GPUs or CUDA installed, treat the CUDA version in the url as cpu. See https://github.com/rusty1s/pytorch_scatter for more details about installation.

Finally, install TorchDrug via

pip install torchdrug

From Source

git clone https://github.com/DeepGraphLearning/torchdrug
cd torchdrug
pip install -r requirements.txt
python setup.py install

Windows (PowerShell)

For Windows, we need to additionally install Visual Studio to enable the JIT compilation. If you don’t have Visual Studio installed, you can get a minimal version of the build tools for Visual Studio at https://visualstudio.microsoft.com/downloads/.

Note

For non-English systems, you must select English as the language for Visual Studio. Otherwise, the ANSI encoding of Windows will cause errors in Python.

We then setup a command-line environment for JIT compilation. Launch a PowerShell as administrator, and install the following extensions.

Install-Module Pscx -AllowClobber
Install-Module VSSetup

Initialize Visual Studio in PowerShell with the following commands. We need to change the library path based on our own Python path.

Import-VisualStudioVars -Architecture x64
$env:LIB += ";C:\Program Files\Python37\libs"

The above commands should be executed for every PowerShell session. To setup this for all PowerShell sessions, we can write them to the PowerShell profile. The profile can be found by the $profile command in PowerShell. You may need to create the profile if you use it for the first time.

Apple Silicon (M1/M2 Chips)

PyTorch supports Apple silicon from version 1.13. While torch-scatter and torch-cluster don’t have pre-compiled binaries for Apple silicon, we can compile them from their sources. See https://github.com/rusty1s/pytorch_scatter/issues/241 for more details.

pip install torch==1.13.0
pip install git+https://github.com/rusty1s/pytorch_scatter.git
pip install git+https://github.com/rusty1s/pytorch_cluster.git
pip install torchdrug

Note TorchDrug runs on Apple silicon CPUs, but doesn’t support mps devices.