This repository includes the following example scripts:
- MLP: Full-batch MLP training based on paper features and optional Node2Vec features (
--use_node_embedding
). For training with Node2Vec features, this script requires node embeddings be saved inembedding.pt
. To generate them, please runpython node2vec.py
[requirestorch-geometric>=1.5.0
]. - GNN: Full-batch GNN training using either the GCN or GraphSAGE operator (
--use_sage
) [requirestorch-geometric>=1.6.0
].
# Run with default config
python gnn.py
# Run with custom config
python gnn.py --hidden_channels=128
The tsv file that maps paper IDs into their titles and abstracts are available here.
There are three columns: paperid \t title \t abstract.
You can obtain the paper ID for each node at mapping/nodeidx2paperid.csv.gz
of the downloaded dataset directory.