CoDiSum

Code and data of the paper Commit Message Generation for Source Code Changes.

Intro

CoDiSum is a deep neural network model used to generate commit message for source code changes. It take processed diff as input and outputs natural language description.

In this repository, we provide the implementation of our CoDiSum model with Python Keras APIs.

Requirements

Python >= 3.6.2
Numpy >= 1.15.0
Keras >= 2.1.5
Tensorflow >= 1.7.0

Usage

example usage:

Create a dir models to save the model file and run

python CopyNetPlusgen.py

This command will lead to a full training and prediction process. Related parameters are in CopyNetPlusgen.py file, named in uppercase.

Datasets

We put the data set in a compressed file data4CopynetV3.zip, You can extract directly to the current folder.

Model

Citing

If you find CoDiSum useful in your research, we ask that you cite the following paper:

@inproceedings{
author = {Xu, Shengbin and Yao, Yuan and Xu, Feng and Gu, Tianxiao and Tong, Hanghang and Lu, Jian},
title = {Commit Message Generation for Source Code Changes},
booktitle = {Proceedings of the 28th International Joint Conference on Artificial Intelligence},
year = {2019},
}