Most of the code is from: https://github.com/uoguelph-mlrg/GGM-metrics
Python version used is 3.8.10 Most of the requirements are in requirements.txt
pip install -r requirements.txt
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv\
torch-geometric -f https://data.pyg.org/whl/torch-1.11.0+cpu.html
Requirements are for testing on CPU. For testing on GPU, please install cuda version of the libraries.
You can download the data using download_datasets.sh, (run sh download_datasets.sh in terminal).
For mixing with GRAN generated graphs you should also download GRAN graphs through download_gran_graphs.sh, (run sh download_gran_graphs.sh in terminal).
The arguments used for training is provided in config.py.
A general expression of how you can run an experiment is as:
python main.py --permutation_type={permutation type} {feature_extractor}\
--dataset={dataset} {other args if you want to change default}
Permutation type can be one of the {mixing-random, rewiring-edges, mode-collapse, mode-dropping}. Feature extractor can be graphcl, infograph, or random-gin. If the network is not already pretrained, code will train the model, save it in the saved_models, and run the experiments with the trained model.
For running with the hyperparameters used in the paper you can make --use_predefined_hyperparams equal to true. Making this argument true will ignore the num_layers, epochs, hidden_dim set in the arguments and replace them with the default one. For checking other changeable arguments check config.py
A simple experiment running example is like this:
python main.py --permutation_type=rewiring-edges --feature_extractor=graphcl \
--dataset=grid --num_layers=2 --hidden_dim=16
Rank correlations are automatically computed and printed at the end of each experiment, and results are stored in experiment_results/.
You can also use create_bash_script.py for creating a set of experiments. After running it, bash commands will be saved in all_commands.sh. You can run it through the terminal with command sh all_commands.sh