search

mims-harvard / SubGNN

Subgraph Neural Networks (NeurIPS 2020)
https://zitniklab.hms.harvard.edu/projects/SubGNN
MIT License
185 stars 33 forks source link

Data preprocessing and conda config file #5

Closed rmwu closed 3 years ago

rmwu commented 3 years ago

Hi! Thank you for all your documentation on this project. :) It's been very helpful so far! I'm trying to preprocess a custom dataset using the scripts provided, and I'm running into issues with dependencies.

I created the conda environment with the yml provided, but line 201 "torch==1.0.0" seemed to override line 68 "pytorch=1.4.0=py3.7_cuda10.1.243_cudnn7.6.3_0" so my installation was for 1.0.0. To obtain the correct version, I manually installed version 1.4.0, but that didn't seem to include the right version of torch-geometric, as GraphSAINTRandomWalkSampler appears in a later version. Would it be possible to update the environment yml with a pruned version of the requirements, or is there another sampler you would recommend?

EmilyAlsentzer commented 3 years ago

Hi @rmwu, you're right that the GraphSaint sampler requires a newer version of pytorch geometric. We used two different environments for GIN vs GCN-GraphSaint embeddings because pytorch geometric was updated mid-project, but I just updated everything to the new version of Pytorch Geometric. On the modeling side, the only difference is refactoring of the functions that were used in the propagate function in subgraph_mpn.py. There's a new SubGNN.yml file that reflects the updated environment with the new version of pytorch geometric.

EmilyAlsentzer commented 3 years ago

Closing now, but feel free to reopen if you have any more questions.

rmwu commented 3 years ago

Thank you!!! This was very helpful! (Didn't realize you left it open :) )