Codes for COLING 2022

This repository contains the official codes for our paper at COLING 2022: Semantic-Preserving Adversarial Code Comprehension.

Overview

We conduct our experiments on three datasets: Defects4J for Defect Detection, CodeSearchNet for Natural Language Code Search and CodeQA for Question Answering over Source Code.

You can find codes and instructions in each folder corresponds to each dataset:

Defect Detection

Natural Language Code Search

Question Answering over Code

Dependencies and Environment

To install the dependencies, please run:

pip install -r requirements.txt

Besides, we conduct our experiments on the following environment:

torch: 1.10.2
python: 3.7.9
CUDA Version: 11.4
GPU: RTX 3090 24G

We strongly recommand that you run the experiments on the same environment to ensure the reproductivity.

Citation

If you find our paper and repository useful, please cite us in your paper:

@inproceedings{li-etal-2022-semantic,
    title = "Semantic-Preserving Adversarial Code Comprehension",
    author = "Li, Yiyang  and
      Wu, Hongqiu  and
      Zhao, Hai",
    booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "International Committee on Computational Linguistics",
    url = "https://aclanthology.org/2022.coling-1.267",
    pages = "3017--3028",
}