Science Score: 13.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
✓codemeta.json file
Found codemeta.json file -
○.zenodo.json file
-
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (7.1%) to scientific vocabulary
Repository
Basic Info
- Host: GitHub
- Owner: PKU-DAIR
- Default Branch: main
- Size: 5.55 MB
Statistics
- Stars: 7
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
An Experimental Evaluation of Using Deep Convolution on Graph-Structured Data
Requirements
Environments: Xeon Gold 5120 (CPU), 384GB(RAM), TITAN RTX (GPU), Ubuntu 16.04 (OS).
The PyTorch version we use is torch 1.7.1+cu110. Please refer to the official website -- https://pytorch.org/get-started/locally/ -- for the detailed installation instructions.
To install all the requirements:
setup
pip install -r requirements.txt
Experimental Analysis
We implement ResGCN, DenseGCN, MLP+Res, MLP+Dense, SGC, and 2 GCN variants on our own in ./src/models.py:
the code of ResGCN and DenseGCN is in ./src/gcn_sc.py;
the code of MLP+Res and MLP+Dense is in ./src/mlp_sc.py;
the code of SGC is in ./src/sgc.py;
the code of GCN with $Dt=2$ and GCN with $Dp=2Dt$ is in ./src/gcn2dt.py and ./src/gcn_dp2dt.py, respectively.
the code for printing the gradient of the first layer of GCN is in ./src/print_gradient.py;
the code for the scalability experiment is provided in ./src/scalability/, please run gengraph.py first to generate artificial graphs; then run appnp/gcn/dgmlp.py --n="graphsize", where "graph size" varies from 100,000 to 1,000,000 with the step of 100,000.
We also provide the official code of DAGNN, S$^2$GC, and Grand under ./src/.
DGMLP Training
To test the performance of DGMLP on the Cora, Citeseer, Pubmed dataset, please run this command:
train
bash ./src/run.sh
Node Classification Results:
Owner
- Name: DAIR Lab
- Login: PKU-DAIR
- Kind: organization
- Email: bin.cui@pku.edu.cn
- Location: Beijing
- Repositories: 4
- Profile: https://github.com/PKU-DAIR
Data and Intelligence Research (DAIR) Lab @ Peking University