---
arxiv: 1807.0432
dataset_info:
  features:
  - name: functionSource
    dtype: string
  - name: CWE-119
    dtype: bool
  - name: CWE-120
    dtype: bool
  - name: CWE-469
    dtype: bool
  - name: CWE-476
    dtype: bool
  - name: CWE-other
    dtype: bool
  - name: combine
    dtype: int64
  splits:
  - name: train
    num_bytes: 832092463
    num_examples: 1019471
  - name: validation
    num_bytes: 104260416
    num_examples: 127476
  - name: test
    num_bytes: 104097361
    num_examples: 127419
  download_size: 535360739
  dataset_size: 1040450240
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- text-classification
tags:
- code
---

This is an unofficial HuggingFace version of "[Draper VDISC Dataset - Vulnerability Detection in Source Code](https://osf.io/d45bw/)" dataset from "[Automated Vulnerability Detection in Source Code Using Deep Representation Learning](https://arxiv.org/abs/1807.04320)".

***

Draper VDISC Dataset - Vulnerability Detection in Source Code

The dataset consists of the source code of 1.27 million functions mined from open source software, labeled by static analysis for potential vulnerabilities. For more details on the dataset and benchmark results, see https://arxiv.org/abs/1807.04320.

The data is provided in three HDF5 files corresponding to an 80:10:10 train/validate/test split, matching the splits used in our paper. The combined file size is roughly 1 GB. Each function's raw source code, starting from the function name, is stored as a variable-length UTF-8 string. Five binary 'vulnerability' labels are provided for each function, corresponding to the four most common CWEs in our data plus all others:

```
CWE-120 (3.7% of functions)
CWE-119 (1.9% of functions)
CWE-469 (0.95% of functions)
CWE-476 (0.21% of functions)
CWE-other (2.7% of functions)
```

Functions may have more than one detected CWE each.

Please cite our paper if you use this dataset in a publication: https://arxiv.org/abs/1807.04320

This project was sponsored by the Air Force Research Laboratory (AFRL) as part of the DARPA MUSE (https://www.darpa.mil/program/mining-and-understanding-software-enclaves) program.

About Draper (https://www.draper.com) - Draper is an independent, not-for-profit corporation, which means its primary commitment is to the success of customers' missions rather than to shareholders. For either government or private sector customers, Draper leverages its deep experience and innovative thinking to be an effective engineering research and development partner, designing solutions or objectively evaluating the ideas or products of others. Draper will partner with other organizations — from large for-profit prime contractors, to government agencies, to university researchers — in a variety of capacities. Services Draper provides range from concept development through delivered solution and lifecycle support. Draper's multidisciplinary teams of engineers and scientists can deliver useful solutions to even the most critical problems.