search

VITA-Group / graph_ladling

[ICML2023] Graph Ladling: Shockingly Simple Parallel GNN Training without Intermediate Communication. Ajay Jaiswal, Shiwei Liu, Tianlong Chen, Ying Ding, and Zhangyang Wang
MIT License
11 stars 1 forks source link

Missing document #1

Open 666psj666 opened 9 months ago

666psj666 commented 9 months ago

Hi author, your work is excellent, but I'm having some problems trying to run the code. I want to know what is./cpp_extension/sample.cpp in the file directory? Where do I get it? The./cpp_extension/sample.cpp file cannot be found when running the GraphSampling/setup.py program

yanweiyue commented 3 months ago

the same problem

aomisai commented 2 weeks ago

@666psj666 @yanweiyue I tried making cpp_extension directory in GraphSampling/ and asked chatgpt for the missing code. I added it and ran setup.py, which worked. Not sure how legit is but at least it let me run the code. Though there are many many issues still...

`#include <torch/extension.h>

// A simple placeholder function for edge sampling torch::Tensor edge_sample2(torch::Tensor p_cumsum, int64_t batch_size, torch::Tensor _null) { return torch::randint(0, p_cumsum.size(0), {batch_size}); }

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) { m.def("edge_sample2", &edge_sample2, "A placeholder edge sampling function"); }`

aomisai commented 3 days ago

I decided to fix this project on my own. Since the original code doesnt provide certain code to recreate the results from the paper, I'm developing the missing files and features. Feel free to visit my fork.