This repository provides the implementation of the graph models proposed in our paper.

The key idea of GINN is to streamline the message-passing procedure based on intervals of control flow graphs. In that sense, the design can be seamlessly incorporated into any graph-based models for learning embeddings of source code. We give two specific instances: GINN and GINN-based RNN Sandwich.

• Python 3

• pip install -r requirements.txt

1. GGNN-based model: in models/GGNN.py

2. GINN model: in models/GINN.py

We have provided sample files to test both models (in data). The format of the sample files are explained in data/README.md

Give command like below for running the models:

cd models
python GGNN.py --config-file config/config-simple-AST.json
python GINN.py --config-file config/config-simple-interval.json

Here are the screenshots of the two models.

We provide an interval generator here. The generator parses Java files to CFGs, ASTs and intervals by Spoon.

Change directories into the one with Dockerfile, run the folllowing commands, and follows command in Test the models:

docker build --tag ginn:1.0 .
docker run -it ginn:1.0 /bin/bash