VeloAE
VeloAE copied to clipboard
some problem about PyTorch and torch-geometric version
The project is interesting and paper is enlightening for me.
- The setup.py of this project is not very clear about the versions of pytorch and torch-geometric. There are some big differences between pytorch 1.7, 1.8, 1.9 and 1.10, and there are obvious differences between torch-geometric versions 1 and 2.Default pip installation is newest version of pytorch 1.11-cpu and torch-geometric 2.0.4, which leads to my installation error. Can you tell me the specific version of your pytorch and torch-geometric version.
- If you build the project by CONDA, can you use CONDA by command “env export > veloae.yaml” to export the specific version files of all packages and share them with me? That would be better and I greately appreciate it.
Thank you for your interest in this study. Indeed, the latest dependencies of pytorch_geometric are somewhat brittle,
Better install pytorch and pyG independently. Please see the below cmdline history creating a conda environment in linux, which works smoothly on my server.
837 conda create -n veloAE
838 conda activate veloAE
839 git clone https://github.com/qiaochen/VeloAE.git
840 cd VeloAE
841 conda install python=3.7
842 pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
843 python (can skip, I was checking installation)
844 pip install torch-scatter -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
845 pip install torch-sparse -f https://data.pyg.org/whl/torch-1.11.0+cu113.html
846 pip install torch-geometric
847 python (can skip, I was checking installation)
848 pwd (make sure we are in project root folder)
849 pip install .
A yam file for this new environment is also uploaded to the project root.