ADG-Seq2Seq: Embedding API Dependency Graph for neural code generation

An Encoder_Decoder_Embedder based model

Our paper:Embedding API Dependency Graph for neural code generation

---

Requirements

• Python 3.7
• Pytorch 1.5.0
• CUDA 1.5.0

---

Quick Start

Step 1: Train a new model

To train a new model, you can run file training.py like:

    python3 training.py [HS/MTG/EJDT]

the models will be stored into /model/*.

• Since large files are not easy to upload and download on github, we put the complete data set on google drive, click here for downloading. We also put some portions of dataset as demos, so before running, please rename them (if use).
• Before training, please new a folder ./model to store models.

Step 2: Evaluation

You can run the model by:

    python3 eval.py [HS/MTG/EJDT]

---

Examples

Code Description:

Reads the contents of the specified block into the buffer's page. If the buffer was dirty, then the contents of the previous page are first written to disk.

Example Code:

void assignToBlock(BlockId blk){
    internalLock.writeLock().lock();
    try {
        this.blk=blk;
        contents.read(blk);
        lastLsn=readPage(contents, LAST_LSN_OFFSET);
    }
    finally {
        internalLock.writeLock().unlock();
    }
}

---

Citation:
If you find this code useful in your research, please cite our paper:

@article{lyu2021embedding,
  title={Embedding API dependency graph for neural code generation},
  author={Lyu, Chen and Wang, Ruyun and Zhang, Hongyu and Zhang, Hanwen and Hu, Songlin},
  journal={Empirical Software Engineering},
  volume={26},
  number={4},
  pages={1--51},
  year={2021},
  publisher={Springer}
}