Prerequisites: torch, torch-geometric
Example commands to run a single model configuration,
python main.py --data=PROTEINS --lr=0.1
python main.py --data=DHFR --use_node_attr --ignore_node_labels
python main.py --data=COLLAB --aggregation=Mean --lr=0.001
To run a number of configurations, to use for model selection later, set the required configurations in config.txt and run:
python main.py --use_config
Once the pickle file is generated for outputs of all configurations, use model selection to select models. If you use "--aggergation=both", make sure that both files GIN_MMD_<dataset>_<seed>.pkl and GIN_Mean_<dataset>_<seed>.pkl have been generated in the output folder. Here are two examples:
python model_selection.py --data=PROTEINS --seed=1213 --aggregation=MMD
python model_selection.py --data=AIDS --aggregation=both
Run both files with --help to see all parameters that can be input.
Here is a chunk of code which should run all experiments needed to recreate results in GLAM:
python main.py --data=PROTEINS --use_config --config_file=config/MMD_config.txt
python main.py --data=PROTEINS --aggregation=Mean --use_config --config_file=config/Mean_config.txt
python model_selection.py --data=PROTEINS