This library turns java source code (.java files) into Augmented ASTs (.gml (graphml) files) as per the paper Open Vocabulary Learning on Source Code with a Graph-Structured Cache.

More specifically, you list the names of any java repos from the Maven Repository that you'd like to convert into a dataset, and then this library will automatically download those repos and generate Augmented ASTs of all their constituent files, one .gml file per .java file.

You'll need Apache Maven installed. (And the basic linux command line utilities.)

Then run

cd <root directory of this repo>
mvn install -DskipTests
cd <root directory of this repo>/javaparser-dloc
mvn install -DskipTests

There is a file called repositories.txt in javaparser-dloc/scripts. You should change this file to contain whatever repo names from the Maven Repository that you'd like to process into datapoints. The format is one repo per line, each line reading <org name>:<repo name>:<version number>. At the moment, repositories.txt contains the names of the 18 Maven repos used in the Deep Learning On Code With A Graph Vocabulary.

Once you've edited repositories.txt, run the createDatasets.sh script as follows:

export dataset=<path to where you'd like the dataset to go>
cat repositories.txt | xargs -I{} <root directory of this repo>/javaparser-dloc/scripts/createDatasets.sh {} $dataset

Now that you've downloaded and built the repos, process them all into graphml-formatted files:

ls $dataset | xargs -I{} <root directory of this repo>/javaparser-dloc/scripts/processDataset.sh {} $dataset

Feel free to get in touch with Milan Cvitkovic or any of the other paper authors. We'd love to hear from you!