DeepLineDP: Towards a Deep Learning Approach for Line-Level Defect Prediction

Reproduction of C. Pornprasit; C. Kla Tantithamthavorn, DeepLineDP: Towards a Deep Learning Approach for Line-Level Defect Prediction (2023).

📑 Article

🌿 Overleaf

Team organisation:

🛠️ Board

📦 Drive

📄 Team policy

📄 Leader schedule

📄 Team expectations agreement

Similar articles:

LineDP: Predicting Defective Lines Using a Model-Agnostic Technique

Authors

• Jakub Tkaczyk
• Karol Waliszewski
• Mikołaj Macioszczyk

Reproduction

Steps to replicate the study:

1. Access the shared file 🛠️ Main - with our research - DeepLineDp.
2. Create a copy of that file: File -> Save a copy in Drive.
3. On your copy of Main - with our research - DeepLineDP, execute all the tasks. They include data preprocessing, training DeepLineDP, RandomForest, XGBoost, LighGBM models, predicting results, top-k tokens investigation and comparison of the classifiers based on MCC, BA metrics. Predictions are stored as .csv files (one file per release of the library) located in ./comparisonModelsExperiment/output, split by model folders: prediction/DeepLineDP, RF-line-level-result, XGB-line-level-result, LGBM-line-level-result. The comparison results of the classifiers with metrics are located in ./comparisonModelsExperiment/output/figure.
4. The workflow asks you to connect a google drive. This is optional but will provide better user experience - it's possible to download all of the results and figures.
5. The next step is the evaluation results: Recall@Top20LOC, Effort@Top20Recall and IFA. After that the top-k investigation will be performed and at the end MCC, BA metrics will be calculated.
6. The ./comparisonModelsExperiment/output folder will be copied to your google drive as comparisonModelsExperimentFigure.zip. You can then download the file from google drive and unzip on your local machine.
7. The evaluation results are located in ./figure folder.