Empirical Study of Just-In-Time Defect Prediction

Replication package for the paper entitled: Revisiting Pre-trained Models and Feature Fusion Strategies for Just-In-Time Defect Prediction

Data Preparation

The Original and Unified datasets can be found here. Download *.tar.gz and extract it under the Dataset/fine-tuning/JITDefectPrediction/ folder via tar -zxvf *.tar.gz.

Pretrained model files Preparation

Manually download the added_tokens.json, config.json, merges.txt, pytorch_model.bin, special_tokens_map.json, tokenizer_config.json, and vocab.json from Hugging Face-CodeT5, upload them to models/codet5_base/.

Environment Settings

• GPU: Nvidia Tesla P40
• OS: CentOS 7.6

git clone https://github.com/Hugo-Liang/EmS-JIT-DP.git
cd EmS-JIT-DP
conda create -n EmS python=3.8
conda activate EmS
pip install torch==2.0.0+cu117 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

Run

1. CodeT5+RF: CodeT5 with random forest

• 1.1 CodeT5 with semantic feature

bash scripts/finetune_jitdp_SF.sh -g 0

• 1.2. Random Forest with expert features

python RF.py

• 1.3. Average prediction results from CodeT5 and Random Forest

python combine.py

2. CodeT5+EF: CodeT5 with concatenated embeddings

• CodeT5 with semantic feature and expert features

bash scripts/finetune_jitdp_SF_EF.sh -g 0

Get Involved

Please create a GitHub issue if you have any questions, suggestions, requests or bug-reports.