BiGraph

BiGraph is a Python package for Link prediction in bipartite networks.

• Bug reports: https://github.com/bi-graph/bigraph/issues

Node based similarities and Katz has been implemented. you can find algorithms in bigraph module. Algorithms implemented so far:

$ pip install bigraph

Simple example

Predicting new links in a randomly generated graph using Adamic-Adar algorithm:

from bigraph.predict import aa_predict
from bigraph.preprocessing import import_files, make_graph


def adamic_adar_prediction():
    """
    Link prediction on bipartite networks
    :return: A dictionary containing predicted links
    """

    df, df_nodes = import_files()
    print(df)
    print(f"Graph Nodes: ", df_nodes)
    G = make_graph(df)
    print(G)
    predicted = aa_predict(G)  # Here we have called Adamic Adar method from bigraph module
    return predicted


# Executing the function

if __name__ == '__main__':
    adamic_adar_prediction()

Evaluating Adamic-Adar algorithm.
You can try other provided prediction algorithms by replacing the "aa" argument.

from bigraph.evaluation.evaluation import evaluate
from bigraph.preprocessing import import_files, make_graph


def adamic_adar_evaluation():
    """
    Evaluate Adamic-Adar algorithm using 10-Fold cross-validation 
    :return: A dictionary containing the evaluation results
    """
    df, df_nodes = import_files()
    G = make_graph(df)
    results = evaluate(G, k=10,
                       method='aa')  # Here we have evaluated adamic-adar
    # methods using evaluation module. Methods are 'jc', 'aa', 'pa', 'cn'
    return results


# Executing the function
if __name__ == '__main__':
    adamic_adar_evaluation()

Metrics

Metrics that are calculated during evaluation:

• For further usages and calculating different metrics

Dataset format

Your dataset should be in the following format (Exclude the 'Row' column):

• Note that running

  from bigraph.preprocessing import import_files
  df, df_nodes = import_files()

  will create a sample graph for you and will place it in the inputs directory.
• Although the weight has not been involved in current version, but, the format will be the same.

More examples

Predicting new links in a randomly generated graph using following algorithms:

• Preferential attachment
• Jaccard similarity
• Common neighbours

from bigraph.predict import pa_predict, jc_predict, cn_predict
from bigraph.preprocessing import import_files, make_graph


def main():
    """
    Link prediction on bipartite networks
    :return:
    """
    df, df_nodes = import_files()
    G = make_graph(df)
    pa_predict(G)  # Preferential attachment
    jc_predict(G)  # Jaccard coefficient
    cn_predict(G)  # Common neighbors


# Executing the function
if __name__ == '__main__':
    main()

References

Future work

• [x] Modulate the functions
• [ ] Add more algorithms
• [ ] Run on CUDA cores
• [ ] Make it faster using vectorization etc.
• [ ] Add more preprocessors
• [ ] Add dataset, graph, and dataframe manipulations
• [ ] Unify and reconstruct the architecture and eliminate redundancy

Notes

• It can export the graph in .json and .gexf format for further usages. For instance: Gephi etc.

If you found it helpful, please give us a :star:

License

Released under the BSD license

Copyright © 2017-2021 BiGraph Developers
Soran Ghadri ([email protected])
Taleb Zarhesh ([email protected])