The code to the paper Edge-Splitting MLP: Node Classification on Homophilic and Heterophilic Graphs without Message Passing

Requirements

Install required packages with: pip install -r requirements.txt

Structure

The start point of the code is main.py. The dataset is constructed in load_dataset.py. Models can be found in models folder.

Running the Code

To run the code there are several paramters which can be given along. Available models are: esmlp, gcn, mlp, gcn, sage, graphmlp, esgnn, linkx Possible paramters for different settings are: --data,--hidden, --epochs, --alpha, --beta, --order, --re_eps, --ir_eps.

To get the results of ES-MLP run for example: Cora: python main.py --model="esmlp" --data="cora" --epochs=300 alpha=1 --beta=0.001 --order=2 --setting_number=10000 Roman-empire: python main.py --model="esmlp" --data"Roman-empire" alpha=1 --beta=0.00001 --order=2 --setting_number=10000

Results

The single runs are stored in Results folder

Other experimemnts

Robustness Analysis Experiment can be run with: python experiment_edge_noise.py --model="esmlp" --data="Cora" --distribution="Uniform"