Investigating the Interplay between Features and Structures in Graph Learning

This is the repo of the paper:

Investigating the Interplay between Features and Structures in Graph Learning, Daniele Castellana and Federico Errica, 20th MLG workshop @ ECML-PKDD

Run the experiments

In our experiments, we assess 5 models on 6 synthetic datasets.

Models:

• MLP
• GCN
• GATv2
• GraphSAGE
• PNA

Datasets:

• N1-most-common-neighbours-type
• N2-least-common-neighbours-type
• N3-parity-neighbours-type
• S1-multipartite-easy
• S2-multipartite-random
• S3-count-triangles-balanced

To execute a single experiment it is enough to run the following command: python run.py --config-file config/MODEL_NAME/DATASET_NAME.yaml --results-dir YOUR_RESULTS_DIR --num-workers NUM_WORKERS

Note that this command executes the models selection in a parallel way. To execute it sequentially, you should set NUM_WORKERS=1. See the config file to check the hyperparameters values validated.

Run all the experiments

The script run_all.sh allows to execute all the experiments of our paper (all the models on all the datasets). The script is executed by the following command:

.\run_all.sh YOUR_RESULTS_DIR NUM_WORKERS

The two arguments are mandatory.