PyTester: Deep Reinforcement Learning for Text-to-Testcase Generation 🪄

⚙️ Installation

pip install -r requirements.txt

🏃 Running

Please check the path of dataset and output directory before running.

Finetuning

This script is for the policy training step in the paper. The model in this step is the CodeT5 baseline (see baselines/Finetune CodeT5).

python finetuning.py

RL Training

This script is for the policy optimizing step in the paper. The model in this step is our PyTester (see results in the folder results).

python ppo_training.py

Inference

This inference script is used for CodeT5 and PyTester models only. For other baseline models, please checkout the folder baselines.

python inference.py

Evaluation

This evaluation script applies to all the models.

python evaluation.py

💾 Datasets and Models

• The processed APPS dataset (train and test sets) and the PyTester model can be accessed here.

• The MBPP and HumanEval datasets are in the folder datasets.

• The baseline result logs and predictions are in the folder baselines.

Reference

If you use our code or PyTester, please cite our PyTester Paper.

@article{takerngsaksiri2024tdd,
  title={TDD Without Tears: Towards Test Case Generation from Requirements through Deep Reinforcement Learning},
  author={Takerngsaksiri, Wannita and Charakorn, Rujikorn and Tantithamthavorn, Chakkrit and Li, Yuan-Fang},
  journal={arXiv preprint arXiv:2401.07576},
  year={2024}
}