This is the online repository of the OOPSLA 2023 paper titled "Two Birds with One Stone: Boosting Code Generation and Code Search via a Generative Adversarial Network". We have released the source code and dataset. The doi of our artifact is: https://doi.org/10.5281/zenodo.7824776.
- Dataset: Our evaluation is based on the large-scale CodeSearchNet dataset. Use the following command to download and preprocess the data:
cd dataset
bash run.sh
cd ..- Dependencies:
pip install -r requirements.txt
[Optional] We have built the tree-sitter parser stored at evaluator/CodeBLEU/parser/languages.so. If it doesn't work for you, it can be rebuilt with the following command:
cd evaluator/CodeBLEU/parser
bash build.sh- Training
bash sh/train.sh [python/java] [CodeT5/Natgen] [CodeBERT/GraphCodeBERT]- Evaluation Evaluate generator:
bash sh/evaluate.sh [python/java] [CodeT5/Natgen]Evaluate discriminator:
We evaluate the discriminator by reusing the code from CodeBERT and GraphCodeBERT. According to the Evaluate section in the corresponding model's Readme, replace model_name_or_path with the respective discriminator that you want to evaluate.