函数级漏洞检测

【实例简介】

open-source project for source code-level vulnerability detection based on the supervised machine learning technique. The project implementation is based on the paper: [Deep Learning-Based Vulnerable Function Detection: A Benchmark]

【实例截图】

## Instructions & Usage
Unzip the zip file of this repository, one will see the following folders:
* The config folder -- containing the configuration file.
* The data folder -- containing the source code functions (vulnerable and non-vulnerable).
* The result folder -- containing the sample results.
* The src folder -- containing the code for model training and test.

And there are two Python script files:
* main.py -- for training and testing a specified network model. By specifying different options/parameters, users can apply different embedding methods and switch between training and testing mode.
* Obtain_representations.py -- for obtaining high-level representations from a trained network model.
【核心代码】

| Options | Description                                                                                   |
|---------|-----------------------------------------------------------------------------------------------|
| config  | Path to the configuration file.                                                                        |
| seed    | Random seed for reproduction of the results.                                       |
| data_dir    | The path of the code base for training. (can be obtained by download & unzip the files under data folder. By default, it is `data/`.) |
| logdir  | Path to store training logs (log files for Tensorboard). By default, it is `logs/`                                                   |
| output_dir  | The output path of the trained network model. By default, it is `result/models/<model_name.h5>`                                                |
| trained_model   | The path of the trained model for test. By default, the trained models are stored in `result/models/`                                                      |                                                               
| embedding |  The embedding method for converting source code sequences to meaningful vector representations. Currently, we also support Word2vec, GloVe and FastText. By default, the Word2vec method is used. |
| test   | Switch to the test mode.                                                               |
| verbose    | Show all messages.