If you are already familiary with running autogluon for node classification, i recommend you to start from AG.py
You need to prepare your dataset, including graph, labels, train_idx, val_idx, test_idx (you can refer to Line 227 to Line 230, downloading OGB-arxiv dataset)
For train function in Line 234, including two parts, the first part is exactly same with AutoGluon and the second part is about Correct and Smooth.
To implement the code python AG.py
There should be at least X.csv (node features), y.csv (target labels), graph.graphml (graph in graphml format).
You can also have cat_features.txt specifying names of categorical columns.
You can also have masks.json specifying train/val/test splits.
Then run the command:
python run.py --datasets your_datasets --task regression --X_lam 20.0 --X_step 5 --y_lam 2.0 --y_step 5 --lr 0.1 --label_smooth --error_smooth
(Very simple example on our dataset House: zip datasets and then run python run.py --datasets house --task regression --X_lam 20.0 --X_step 5 --y_lam 2.0 --y_step 5 --lr 0.1 --label_smooth --error_smooth)