WinstonDeng/LinkPrediction

Link Prediction and Graph Reconstruction

This is a personal implementation for the task of Link Prediction.

Dependencies

    python3
    pytorch 1.5+
    numpy
    pandas
    linkpred
    dgl

Quickly start

Predict by optimized_based method

    python optimized_based.py

For using different configs, you can modify global args in python file head.

    A_path = './A_similarity.csv' # self matrix A csv path
    B_path = './B_similarity.csv' # self matrix B csv path
    A_B_path = './A_B_adjacent.csv' # correlation matrix between A and B csv path
    hidden_edges_num = 4 # 4,8,16,32,64 
    algorithm='SimRank' # algorithm for modeling graph

Predict by learning_based method

    python learning_based.py

For using different configs, you can modify global args in python file head.

    A_path = './A_similarity.csv' # self matrix A csv path
    B_path = './B_similarity.csv' # self matrix B csv path
    A_B_path = './A_B_adjacent.csv' # correlation matrix between A and B csv path
    hidden_edges_num = 4 # 4,8,16,32,64 
   
    embed_dim = 256 # node embedding size for input
    mid_dim = 1024 # node feature size for gcn hidden layer 
    out_dim = 512 # node feature size for output

    max_epoch = 200 # max training epooch
    neg_node_num = 16 # neg node sample num
    rebuild_graph = False # use for graph reconstruction task
    avg_cnt = 3 # repeat number for avgeraging result
    device=torch.device("cpu") # using cpu or gpu

Todo

fix bugs in optimized_based.py
fix bugs in learning_baesd.py
clean and smiplify code for running faster

License

This project is released under the Apache 2.0 license. Please see the LICENSE file for more information.

Acknowledgement

I would like to thank the linkpred and the DGL.

WinstonDeng / LinkPrediction