This project represents the backend and model code for A-m.
Please use the provided setup.py to install the package and dependencies.
This can be used to train a model used for PR-Issue link prediction as:
filename = 'data/google_guava.json'
with open(filename) as f:
repo = jsonpickle.decode(f.read())
with open(filename[:-len('.json')] + '_truth.json') as f:
truth = jsonpickle.decode(f.read())
print('Loaded repository %s' % repo.name)
Linker = Linker(net_size_in_days=14, min_tok_len=2, undersample_multiplicity=100)
Linker.fit(repo, truth)
print('Trained Random Forest classifier')
out_path = os.path.join(os.getcwd(), 'models', repo.name[1:].translate({ord(c): '_' for c in '\\/'}))
os.makedirs(out_path, exist_ok=True)
Linker.persist_to_disk(out_path)
print('Recorded model to disk')
Or make use of the generate_models.py script in backend.
python backend/generate_models.py <path_to_crawled_repository>
To start the backend, edit config.json to point to the desired models and local gits and start it with:
python backend/backend.py
This should be done before the chrome plug-in may be used.
To use a remote backend, edit the SERVER_ADDR variable from chrome_entry/__main__.py from:
SERVER_ADDR = 'http://localhost:8000'
To the desired server.
The corpora and truth data can be found at and downloaded from here
The crawling code is provided in Prediction/gitScraper.py.
The backend can be used for replication as a standalone or it can be used together with a chrome plug-in. For the latter more details at: https://github.com/PPPI/am-chrome
This work was done for the purpose of the following paper: link
This project assumes Python 3.5 or newer.
jsonpickle # For data storage
scikit-learn # For classifier utility functions, recursive-feature elimination etc.
gensim # For TF.IDF implementation
numpy # For numerical operations
pygithub # For interaction with github
pytz # To handle timestamps and time data
pandas # For results interpretation and output
scikit-garden # For Mondrian Forest implementation