Preprocessing datasets

How to preprocess datasets

• Goto src/ directory

  • For directed graph (ex. filmtrust)

    python create_dataset.py --input filmtrust --directed
    

  • For undirected graph (ex. ppi)

    python create_dataset.py --input ppi
    

• Once the above commands generate data_remove_percent_0.5.pkl, we are done with data preprocessing.

Data format

• edges.txt
  • node1[tab]node2
• data_remove_percent_0.5.pkl
  • A dictionary containing the following keys
    • isDirected: whether the graph is directed, index: mapped node index (dictionary)
    • num_nodes: number of nodes, remove_percent: the percentage of test data
    • train_edges: positive edges in the training data, train_edges_neg: negative edges in the training data
    • test_edges: positive edges in the test data, test_edges_neg: negative edges in the test data

If you want to use your own dataset, make a directory and put edges.txt inside the directory.