Use ASTNN To Predict Method Name

How To Run

0. Machine required：

   • GPU (>= 10GB) (if no, modify train.py and set ASTNN(...use_gpu=False...))
   • 32GB main memory
   • 2GB disk

1. Download dataset from one of links here:

   • https://www.dropbox.com/s/l8o1s7qntq42215/comment_data.zip?dl=0
   • 链接：https://pan.baidu.com/s/1WQaANDrcVUthU0pt4GQwJw
     • 提取码：ctyb

2. Extract data directory to the project root, it should contains data/train.json, data/valid.json and data/test.json

3. Configure environment:

   • python 3.7
   • pytorch 1.3 (newest)
   • pandas
   • gensim
   • javalang (use pip to install)
   • maybe others

4. Run python ./preprocess_data.py to preprocess data, some hours needed.

5. Run python ./train.py to train the model. It will train 50 epoch in default, 2-3 hours pre epoch. Model will be saved per epoch.

Hint: if you want to run xxx.py background and print to a file (as a example, xxx.log), use the following command on Linux: nohup python -u xxx.py > xxx.log 2>&1 &