pycode2seq 0.0.1

Inference and training for multiple languages of code2seq

pycode2seq

Training and inference with multiple languages of PyTorch's implementation of code2seq model.

Installation

python setup.py install

Inference

Minimal code example:

import sys
from pycode2seq import DefaultModelRunner

def main(argv):
    runner = DefaultModelRunner(
        save_path = "./tmp",
    )

    #List of embeddings for each method
    method_embeddings = runner.run_embeddings_on_file(argv[1], "kt") 

    #Code2seq predictions
    predictions = runner.run_on_file(argv[1], "kt")

    #Predicted method names
    names = [runner.prediction_to_text(prediction) for prediction in predictions]

if __name__ == "__main__":
    main(sys.argv)

Training

Download astminer and run:

./gradelw shadowJar

Mine projects for paths:

python training/mine_projects.py <data folder> <output folder> <path to astminer's cli.sh>

Combine mined paths:

python training/astminer_to_code2seq.py <data folder/holdout> <output folder> <holdout>

Build vocabulary with build_vocabulary.py from code2seq module

Combine vocabularies:

python training/combine_vocabularies.py

Expand weights:

python training/expand_weights.py