GNN-CONTEXT: A GNN-based Code Context Prediction Approach for Software Development Task

Requirements

Python 3.9+
pip install -r requirements.txt # install the packages

Reproduce the results

1. Dataset formation (dataset_formation/)

   • run the run.sh in the directory to generate the expanded code context model and variable collapsed code context model for 1-step/2-step/3-step.
     • --input_dir: dataset path
     • --action [generate | display | clear]
       • generate: generate the graph for d-step
       • display: show the generated graph in the input_dir
       • clear: delete all generated files

2. GNN model training (code_context_model/)

   • Use following command to embed the graph nodes on different embedding models

   python code_context_model/embedding.py --input_dir <dataset_dir> --output_dir <code_embedding_dir>
   

   • Use following command to generate the train, validation, and test dataset

   python code_context_model/build_dataset.py --input_dir <dataset_dir> --embedding_dir <code_embedding_dir> --output_dir <out_dir> --step <step> --embedding_model <emebdding_model>
   

   • Run the shell script run.sh to train the model, please set the arguments in the script

3. Baseline (baseline) We implemented SOTA and node embedding baselines in this directory.

4. Analyis (analyis) Code for Discussion section.