Logic-error-localization-and-correction-with-machine-learning

Overview

This is a sophisticated machine learning system designed to localize and correct logic errors in code. It leverages state-of-the-art techniques, including graph neural networks and pseudocode analysis, to improve the accuracy of automated program repair. This repository includes scripts for data preparation, model training, and evaluation, reflecting the comprehensive methodology discussed in "Logic Error Localization and Correction with Machine Learning."

Getting Started

Prerequisites

• Linux or macOS
• Python 3.6+
• Conda environment

Installation

Clone the repository and navigate to the project directory:

git clone https://github.com/Arrtourz/LogicerrorRepair
cd LogicerrorRepair

Setting Up the Environment

To set up the required environment, follow these steps:

conda create -n DrRepair python=3.6
conda activate DrRepair
# Install required Python packages

Data Preparation

To download and prepare the raw and preprocessed data:

# Download raw data
sh download_raw_data.sh

# Download preprocessed data
sh download_preprocessed_data.sh

Training Models

To train the models on the prepared data:

# Train on DeepFix dataset
sh model/run_deepfix.sh

# Train on SPOC dataset
sh model/run_spoc.sh

Evaluation and Testing

To evaluate the models and start the server for testing:

# Evaluate and test DeepFix model
sh model/server.sh

# Evaluate and test SPOC model
# Similar to above, adjust the script for SPOC evaluation

References

Xu, Z., Sheng, V. S., & Lu, K. (2023). Logic Error Localization and Correction with Machine Learning (Student Abstract). Proceedings of the AAAI Conference on Artificial Intelligence, 37(13), 16372-16373. [https://doi.org/10.1609/aaai.v37i13.27046](https://doi.org/10.1609/aaai.v37i13.27046)

Xu, Z., & Sheng, V. S. (2023, December). Enhancing Logic Error Detection Through Program Pseudocodes. In 2023 30th Asia-Pacific Software Engineering Conference (APSEC) (pp. 643-644). IEEE.

Acknowledgments

This project, LogicerrorRepair, modifies and extends the work originally presented in "DrRepair: Learning to Repair Programs from Error Messages" by Michihiro Yasunaga and Percy Liang. The foundational work introduces a graph-based, self-supervised program repair method leveraging diagnostic feedback, as detailed in their ICML 2020 paper. We express our gratitude for the insights and source code provided by their work, which has significantly influenced the development of our system.

For the original DrRepair project and paper:

• GitHub repository: https://github.com/michiyasunaga/DrRepair
• Citation:

  @InProceedings{Yasunaga20DrRepair,
    author =  {Michihiro Yasunaga and Percy Liang},
    title =   {Graph-based, Self-Supervised Program Repair from Diagnostic Feedback},
    year =    {2020},
    booktitle =   {International Conference on Machine Learning (ICML)},
  }
  

---

This section helps to credit the original authors and their contributions, showing the lineage and evolution of ideas in the field of automated program repair.

License

This project is released under the MIT License. See the LICENSE file for more details.

---