Usage

How to start using DGB package

Dynamic Graph Benchmark: Towards Better Evaluation for Dynamic Link Prediction

1. download from pip

pip install DGB

2. download dataset

import DGB 

data_name = "enron"  #specify the dataset name you want to download here
enron = DGB.TemporalDataSets(data_name="enron")

list all available datasets

data_list = DGB.list_all()
print(data_list) 

# currently supports all datasets from our paper "Towards Better Evaluation for Dynamic Link Prediction"
>>> ["canparl", "contacts", "enron", "flights", "lastfm", "mooc", "reddit", "socialEvo", "UCI", "un_trade", "un_vote", "us_Legis", "wikipedia"]

3. process the dataset into ML ready format and obtain train, val, test split

enron_tgb = enron.process()
train_data = enron_tgb["train"]   
test_data = enron_tgb["test"]
val_data = enron_tgb["validation"]

4. Design your own Temporal Graph Learning Model, or use existing ones

For example methods, please see github

5. Evaluate your method with our provided evaluator

"""
A code snipet to check the evaluator functionality
Please consider this code only as an example, and feel free to modify it according to your requirements
"""

evaluator = DGB.Evaluator()

# y_true & y_pred: ndarray of shape (n_samples,)
# y_true: shows the status of existence of an edge: 0: negative edge, 1: positive edge
# y_pred: the probability of existence of an edge; output of a dynamic graph learning model

# here, we randomly generate two arrays to check the evaluator functionality
np.random.seed(42)
y_true = np.random.randint(2, size=100)
y_pred = np.random.random_sample((100,))

# the evaluation metrics should be in the following list (note: it should be passed as a list)
evaluation_metric_list = ['ap', 'au_roc_score', 'au_pr_score', 'acc', 'prec', 'rec', 'f1']
input_dict = {'y_true': y_true,
              'y_pred': y_pred,
              'eval_metric': evaluation_metric_list
              }
eval_result = evaluator.eval(input_dict, verbose=True)
print(eval_result)

Cite Our Work

If you find the code or datasets useful for your research, please consider citing us:

@inproceedings{poursafaeitowards,
  title={Towards Better Evaluation for Dynamic Link Prediction},
  author={Poursafaei, Farimah and Huang, Andy and Pelrine, Kellin and Rabbany, Reihaneh},
  booktitle={Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
  year={2022}
}

Improve this page