In this repository, we provide lightweight implementation of Devign: Effective Vulnerability Identification by Learning Comprehensive Program Semantics via Graph Neural Networks.

1. Python=3.6
2. Pytorch==1.4.0
3. Deep Graph Library

python main.py \
      --dataset <name_of_the_dataset> \
      --input_dir <directory_of_the_input>;

The input_dir should contain three json files namely

1. train_GGNNinput.json
2. valid_GGNNinput.json
3. test_GGNNinput.json

Each json file should contain a list of json object of the following structure

{
  'node_features': <A list of features representing every nodes in the graph>,
  'graph': <A list of edges>
  'target': <0 or 1 representing the vulnerability>
}

• Let's assume n nodes in the graph are indexed as 0 to n-1. The length of node_features list should be n. Each feature vector should be 100 elements long. Thus the node_features list should be a 2D list of shape (n, 100).

• The length of graph list should be the number of the edges. Each edge should be represented as a three element tuple [source, edge_type, destination]. Where the source and destinations are indices of corresponding node in node_features list. Edge types should be from 0 to max_edge_types.

1. In this implementation, we followed Devign's paper. We could NOT recreate the result in the original paper though.

[1] Zhou, Yaqin, et al. "Devign: Effective vulnerability identification by learning comprehensive program semantics via graph neural networks." arXiv preprint arXiv:1909.03496 (2019).